“In recent history, the basis of telephone company value has been the sharing of scarce resources — wires, switches, etc. – to create premium-priced services. Over the last few years, glass fibers have gotten clearer, lasers are faster and cheaper, and processors have become many orders of magnitude more capable and available. In other words, the scarcity assumption has disappeared, which poses a challenge to the telcos’ “Intelligent Network” model. A new type of open, flexible communications infrastructure, the “Stupid Network,” is poised to deliver increased user control, more innovation, and greater value.”

                     —–Isenberg, D. S., (1998). “The dawn of the stupid network”. ACM netWorker 2, 1, 24-31.

Much has changed since the late 90’s that drove the Telco’s to essentially abandon their drive for supremacy in intelligent services creation, delivery and assurance business and take the back seat in the information services market to manage the ‘stupid network’ that merely carries the information services.  You have to only look at the demise of major R&D companies such as AT&T Bell Labs, Lucent, Nortel, Alcatel and the rise of a new generation of services platforms from Apple, Amazon, Google, Facebook, Twitter, Oracle and Microsoft to notice the sea change that has occurred in a short span of time. The data center has replaced the central office to become the hub from which myriad voice, video and data services are created, and delivered on a global scale. However the management of these services which determines their resiliency, efficiency and scaling is another matter.

While, the data center value has been the sharing of expensive resources – processor speed, memory, network bandwidth, storage capacity, throughput and IOPs – to create premium-priced services, over the last couple of decades, the complexity of the infrastructure and its management has exploded. It is estimated that up to 70% of the total IT budget now goes to the management of infrastructure rather than to develop new services (www.serverdesignsummit.com). It is important to define what TCO (total cost of ownership) we are talking about here because it is often, used to justify different solutions as the following picture showing three different TCO representations of a data center. Figure 1 shows three different TCO views presented by three different speakers in the Server Design Summit in November 2011.  Each graph, while it is accurate, represents a different view. For example, the first view represents the server infrastructure and its management cost. The second one represents the power infrastructure and its management. The third view shows both the server infrastructure and power management. As you can see the total power and its management, while steadily increasing, is only a small fraction of the total infrastructure management cost.  In addition, these views do not even show the network and storage infrastructure and their management. It is also interesting to see the explosion of management cost shown in figure 3 over the last two decades. Automation has certainly improved the number of servers that can be managed by a single person by orders of magnitude. This is borne by the labor cost in the left picture by Intel which shows it is about 13% of the TCO from server view-point. But this does not tell the whole story.

Figure 1: Three different views of Data center TCO presented in the Server Design Summit conference in November 2011 (http://www.serverdesignsummit.com/English/Conference/Proceedings_Chrono.html). These views do not touch the storage, network and application/service management costs both in terms of software systems and labor.

A more revealing picture can be obtained by using the TCO calculator by one of the Virtualization infrastructure vendors. Figure 2 shows percentage Total Cost of Ownership (TCO) (for a 1500 server data center) over five years by each component with and without virtualization.

Figure 2: Five Year TCO of Virtualization According to a Vendor ROI Calculator. While virtualization reduces the TCO from 35% to 25%, it is almost offset by the software, services and training costs.

While virtualization introduces many benefits such as consolidation, multi-tenancy in a physical server, real-time business continuity and elastic scaling of resources to meet wildly fluctuating workloads, it adds another layer of management systems in addition to current computing, network, storage and application management systems. Figure 3 shows a reduction by 50% of the five-year TCO with virtualization. The Virtual Machine density of about 13 allows a great saving in hardware costs which is somewhat off-set by the new software, training and services costs of virtualization.

Figure 3: TCO over 5 Years with virtualization of 1500 servers using 13 VMs per Server. While the infrastructure and administration costs drop, it is almost offset by the software, services and training costs.

In addition, there is the cost of new complexity in optimizing the 13 or so VMs within each server in order to match the resources (network bandwidth, storage capacity, IOPs and throughput) to application workload characteristics, business priorities and latency constraints. According to a storage consultant, Jon Toigo “Consumers need to drive vendors to deliver what they really need, and not what the vendors want to sell them. They need to break with the old ways of architecting storage infrastructure and of purchasing the wrong gear to store their bits: Deploying a “SAN” populated with lots of stovepipe arrays and fabric switches that deliver less than 15% of optimal efficiency per port is a waste of money that bodes ill for companies in the areas of compliance, continuity, and green IT.”

Resource management based data center operations miss an important feature of services/applications management which is that all services are not created equal. They have different latency and throughput requirements. They have different business priorities and different workload characteristics and fluctuations. What works for the goose does not work for the gander. Figure 4 shows a classification of different services based on their throughput and latency requirements presented by Dell in the server design summit. The applications are characterized by their need for throughput, latency and storage capacity. In order to take advantage of the differing priorities and characteristics of the applications, additional layers of services management are introduced which focus on service specific resource management. Various appliance or software based solutions are added to the already complex resource management suites that address server, network and storage to provide service specific optimization. While this approach is well suited for making recurring revenues for vendors, it is not ideally suited for customers to lower the final TCO when all piece-wise TCO’s are added up. Over a period of time, most of these appliances and software end up as shelf-ware while the venodors tout more new TCO reducing solutions. For example, a well known solution vendor makes more annual revenue from maintenance and upgrades than new products or services that help their cutomers really reduce the TCO.

 Figure 4: Various services/Applications characterized by their throughput and latency requirements. Current resource management based data center does not optimally exploit the resources based on application/service priority, workload variations and latency constraints. It is easy to see the inefficiency in deploying a “one size fits all” infrastructure. It will be more eff icient to tailor “dumb” infrastructure and “Stupid Network” pools specialized to cater to different latency and throughput characteristics and let intelligent services provision themselves with the right resources based on their own business priorities, workload characteristics and latency constraints. This requires the visibility and control of service specification, management and execution available at run time which necessitates a search for new computing models.

In addition to the current complexity and cost of resource management to assure service availability, reliability, performance and security, there is even more fundamental issue that plagues the current distributed systems architecture. A distributed transaction that spans multiple servers, networks and storage devices in multiple geographies uses resources that span across multiple data centers. The fault, configuration, accounting, performance and security (FCAPS) of a distributed transaction behavior requires the end-to-end connection management more like telecommunication service spanning distributed resources. Therefore, focusing on only resource management in a data center without the visibility and control of all resources participating in the transaction will not provide assurance of service availability, reliability, performance and security.

Distributed transactions transcend the current stored program control implementation of the Turing machine which is at the heart of the atomic computing element in current computing infrastructure.  The communication is not an integral part of this atomic computing unit in the stored program control implementation of the Turing machine. The distributed transactions require interaction which integrates computing and communication to provide the ability to specify and execute highly temporal and hierarchical event flows. According to Goldin and Wegner, Interactive computation is inherently concurrent, where the computation of interacting agents or processes proceeds in parallel. Hoare, Milner and other founders of concurrency theory have long realized that Turing Machines (TM) do not model all of computation (Wegner and Goldin, 2003). However, when their theory of concurrent systems was first developed in the late ’70s, it was premature to openly challenge TMs as a complete model of computation. Their theory positions interaction as orthogonal to computation, rather than a part of it. By separating interaction from computation, the question whether the models for CCS and the Pi-calculus went beyond Turing Machines and algorithms was avoided. The resulting divide between the theory of computation and concurrency theory runs very deep. The theory of computation views computation as a closed-box transformation of inputs to outputs, completely captured by Turing Machines. By contrast, concurrency theory focuses on the communication aspect of computing systems, which is not captured by Turing Machines – referring both to the communication between computing components in a system, and the communication between the computing system and its environment. As a result of this division of labor, there has been little in common between these fields and their communities of researchers. According to Papadimitriou (Papadimitriou, 1995), such a disconnect within the theory community is a sign of a crisis and a need for a Kuhnian paradigm shift in our discipline.”

Kuhnian paradigm shift or not, a new computing model called DIME computing model (discussed in WETICE2010) provides a convergence of these two disciplines by addressing the computing and the communications in a single computing entity that is a managed Turing machine. The DIME network architecture provides a fractal (recursive) composition scheme to create an FCAPS managed network of DIMEs implementing business workflows as DAGs supporting both hierarchical and temporal event flows. The DIME computing model supports only those computations that can be specified as managed DAGs where a management signaling network overlay allows execution of managed computing tasks (executed by a computing unit called MICE) in each Turing machine node that is endowed with self-management using parallel computing threads. The MICE (see the video referenced in this blog for a description of DIME and its use in distributed computing and its management) constitutes the atomic Turing machine that is controlled by the FCAPS manager in a DIME which allows configuring, executing and managing the MICE to load and execute well specified computing workflow and its FCAPS management. The MICE under parallel real-time control of the DIME FCAPS manager aided by a signaling network overlay provides control over start, stop, read and write abstractions of the Turing machine. Two implementations have proven the existence proof for the DIME network architecture.

Figure 5 shows a DIME network implementing Linux, Apache, MySQL and PHP/Perl/Python web services delivery and assurance infrastructure.

Figure 5: The GUI showing the configuration of a LAMP Cloud (Mikkilineni, Morana, Zito, Di Sano, 2012). Each Apache and DNS are DIME aware running in a DIME aware Linux Operating System which, transforms a process into a managed element in the DIME network. A video describes the implementation of auto-failover, auto-scaling and performance management of the DIME aware LAMP cloud

Look Ma! No Hypervisor or VM in My Cloud (See Video)

The prototype implementations demonstrates a side effect of the DIME network architecture, which combines the computing and communication abstractions at an atomic level, - it decouples the services management from the underlying hardware infrastructure management. This makes it possible to implement highly resilient distributed transactions with auto-scaling, self-repair, state-aware migration, and self-protection – in-short, end-to-end transaction FCAPS management – based on business priorities, workload fluctuations and latency constraints.  No Hypervisors or VMs are required. The intelligent management of services workflow with resilient distributed transactions offers a new architecture for the data center infrastructure. For the first time it will be possible to remove embedding service management in the infrastructure management intelligence using myriad expensive appliances and software systems. It will be possible to design new tiers of dumb infrastructure pools (of servers, storage and network devices) with different latency and throughput characteristics and the services will be able to manage themselves based on policies by requesting appropriate resources based on their specifications. They will be able to self-migrate when quality of service levels are not met. The case for dumb infrastructure on a stupid network with intelligent services management puts forth the following advantages:

1. Separation of concerns: The network, storage and server hardware provides hardware infrastructure management with signaling enabled FCAPS management. They do not encapsulate service management as the current generation equipment does.
2. Specialization: The hardware is designed to meet specific latency and throughput characteristics to simplify its design through specialization. Different hardware with FCAPS management and signaling will provide plug and play components at run time.
3. End-to-end service connection FCAPS management using the signaling network overlay allows dynamic service FCAPS management facilitating self-repair, auto-scaling, self-protection, state-aware migration and end to end transaction security assurance.

Figure 4 shows an example design of a possible storage device using simple storage architecture enabled with FCAPS management over a signaling overlay. It can be easily built with commercially off the shelf (COTS) hardware. This design allows separation of the services management from storage device management and eliminates a host of storage software management systems thus simplifying the data center infrastructure.

Figure 5: A gedanken design of autonomic storage and autonomic storage service deployment using the new DIME network architecture. The signaling overlay and FCAPS management are used to provide dynamic service management. Each service can request, using standard Linux OS services during run time, services from the storage device based on business priorities, workload fluctuations and latency constraints.

It is easy to see that the service connection model eliminates the need for clustering and provides new ways to provide transaction resilience with features such as service call forwarding, service call waiting, data broadcast, 800 service call model etc.

Obviously the new computing model is in its infancy and requires participation from academicians who can validate or reject its theoretical foundation, VCs who can see beyond current approaches and are not satisfied by how many servers can be managed by a single administrator to measure the data center efficiency (as one Silicon Valley VC claimed it as progress in the Server Design Summit) and architects who exploit new paradigms to disrupt the status-quo. The DIME computing model by allowing Linux processes to be converted into a DIME network transcending physical boundaries allows easy migration from current infrastructure to the new one without abandoning legacy applications as the prototype of LAMP cloud demonstrates.

In closing, I like to point out that there have been many calls for a new computing model that combines computing and communication at an atomic computing element level which the Turing machine falls short as discussed above. However, without high bandwidth communication and exploitation of the parallelism that is abundant in the new generation hardware, it is not practically very useful to seriously utilize such new computing models. However, it seems that the hardware advances have outpaced the software advances and perhaps it is about time for computer scientists to seriously take a second look at addressing the software short-fall in dealing with distributed transactions. As the following fable illustrates, it may be futile to look for parallel break-through solutions in a serial boat.

If you have transformational research results, or want to make a real difference in computer science research, see Call for Papers at:

www.workshop.kawaobjects.com and http://WETICE.org

“WETICE 2012 Convergence of Distributed Clouds, Grids and their Management Conference Track is devoted to transform current labor intensive, software/shelf-ware-heavy, and knowledge-professional-services dependent IT management into self-configuring, self-monitoring, self-protecting, self-healing and self-optimizing distributed workflow implementations with end-to-end service management by facilitating the development of a Unified Theory of Computing.”

“Reliability engineering is a bit like alchemy. The field swirls with competing schools of thought. Profound arguments erupt over obscure issues, and there is little consensus on how to proceed even to the extent that we know how to solve many of the hard problems.’’ (K.P. Birman, Reliable Distributed Systems: Technologies, Web Service and Applications (Springer, NY, 2005), p. xix).

With the calls for no less than a Kuhnian paradigm shift in computer science and ensuing controversy over computing models and Turing machines, the current state of the art and science of distributed systems discipline is under scrutiny. With about 70% of Information Technology budget being spent on managing the complexity of distributed systems in an enterprise, the pressure is mounting to seek alternatives.

Here is some food for thought…

Evolution, it seems, assists the systems to strive constantly to improve productivity and optimize their chance of survival. Most of the times, productivity improvements are incremental.  However, occasionally orders of magnitude improvements occur often preceded by a major disruption in the status-quo. When the dust settles, things are never the same as before. In the last sixty years, the Information Technology has seen such changes routinely. I choose COBOL programming, Frame Relays and Asynchronous Transfer Mode Switching (ATM) as examples that demonstrate the changes in computer programming, networking bandwidth and information services switching technologies that have shaped communication, collaboration and commerce on a global scale. In the late 90’s, with billions of lines of entrenched code, COBOL programming and programmers were considered invincible. Today, very few new lines of code are written in COBOL, while still some companies make money on legacy maintenance. The Frame Relay technology provided the high bandwidth communication for businesses and made the Telco reign supreme for a while. The introduction of SONET, optical broadband networks and high-speed Ethernet drastically altered the landscape initiating the demise of Frame Relay technology. Similarly, the ATM switching technology was going to alter the telecommunication world but along the way, came the Internet and IP technology which wiped out a host of high-flying companies in the 90’s.

Today, Virtualization technologies and the Virtual Machine (VM) infrastructure are attempting to create a new wave that is altering how we develop, deploy and operate software based services. With the increase in performance of computing processors, and broadband ubiquity, it is now possible to deploy multiple Operating Systems (OS) in a physical container thus increasing the granularity of multi-tenancy (with multiple Virtual Servers contained in a physical server) of end users with fault, configuration, accounting, performance and security (FCAPS) management isolation for each tenant. The virtualization of the physical server abstraction allows auto-scaling, self-repair, live migration and cloud based deployment of services bringing the advantages of scale and provisioning flexibility.

On the other hand, this introduces a new layer of complexity by bringing the networking and storage management to the virtual server inside a physical server. The introduction of many-cores in a processor creates a hierarchy of networks with different bandwidths now need to be managed to provide end to end transaction FCAPS management. Figure 1 shows the hierarchy of networks.

Figure 1: The hierarchy of networks with different bandwidths requires new management strategies for assuring end-to-end transaction availability, reliability, performance and security.

The VM based service architecture follows current ad-hoc practices of distributed computing and the current discipline of distributed systems with their server-centric and bandwidth limited origins is not ideally suited to exploit the parallelism, and performance of many-core processors, hierarchies of different bandwidth networks and rapidly changing business priorities, workload fluctuations and latency constraints. The lack of resiliency in distributed systems that affects the availability, reliability, performance and security of end-to-end distributed transactions where changes in one component influences other components, traces back to the fundamental computing model that underlies the foundation of computer science and is the basis for current implementation of IT infrastructure which is the von Neumann Stored Program Control (SPC) implementation of the Turing machine.

The Turing machine is an atomic computing unit that according to Church-Turing thesis, “A function is effective computable by a physical system iff it is Turing machine (TM) compatible.” The action of a Turing machine is determined completely by (1) the current state of the machine (2) the symbol in the cell currently being scanned by the head and (3) a table of transition rules, which serve as the “program” for the machine. The von Neumann SPC implementation forms the basis for the origin of computing infrastructure as we know it. Since the first von Neumann implementation, the computing infrastructure has undergone many changes that may or may not be limited by the Turing model. The Turing machine, according to Wegner et al., and Eberbach et al., is a closed world model; dynamic changes to the world outside the TM which occur during the computation have no bearing on the computation itself. The TM’s computation is completely determined by its input, which has to be predefined in advance – it cannot be altered from the outside once the computation begins. TMs compute functions from these inputs to some output; once the output’s value is determined, the computation stops.  Current implementations of operating system controlled process based computations already breach this simple Turing model.  Wegner et al., argue that current services and computer applications that work with unknown or dynamical inputs, and programs like operating systems or database servers that by design never terminate. This is a topic of controversy (Cockshott et al.) and a fertile ground for research on computation models.

Be that as it may, the operating system allows multiple Turing machines (to be precise, von Neumann Stored Program Control (SPC) implementations executing processes or threads) to share the available resources (memory, CPU etc.) based on policies determined external to the computations carried out by the TMs. Therefore, the physical server with an operating system offers signaling and management services to an application to compose an orchestrated workflow as a directed acyclic graph (DAG). The operating system services manage the application processes/threads using the signaling (alerting, addressing, supervision and mediation) and FCAPS management during run time. They instantiate, load and orchestrates a set of Turing machines to execute a computational workflow defined by the application along with other applications based on external policies. However, both the computational workflow defined by the application and its management workflow offered by the operating system are both implemented by the atomic Turing machine computations which are static and closed as discussed above; the workflows have to be predefined at compile time. In spite of this restriction, the flexibility of implementing a computational workflow as a DAG has allowed implementing business workflows where changes in one part of the workflow can influence other parts of the workflow.

While the operating system provides process management in a physical server, the operating system and the physical server itself have evolved to be managed by other applications to provide FCAPS management and optimize availability, reliability, accounting, performance and security management of the system as whole. It is important to note that layers of management are required, albeit hardcoded at compile time of these services, because Gödel’s theorem prohibits self-reflection and TM does not allow dynamic influence from external agents while computation is in progress. The operating system circumvents the self-reflection prohibition by using managed reflection external to the computing units. As von Neumann mentions (von Neumann J (1966) The theory of self-reproducing automata (edited and completed by AW Burks) University of Illinois Press, Urbana, IL. pp. 47, 51-56) “I am a little twisting a logical theorem, but it’s perfectly good logical theorem. It’s a theorem of Gödel that the next logical step, the description of an object, is one class type higher than the object and is therefore asymptotically longer to describe.” In a reply to a question from Burke regarding this comment, Gödel wrote that the theorem mentioned should be “the fact that a complete epistemological description of a language A cannot be given in the same language A.”

The virtual machine technology introduces another layer of management to allow multiple operating systems share the same resources in a physical container thus allowing more flexibility in sharing the resources.  This flexibility comes with a price of complexity. Figure 2 shows current state of affairs in a data center where multiple layers of management in implementing the computing workflows across distributed computing systems. While this architecture successfully executes distributed managed workflows, the resiliency of the end-to-end transaction management suffers from the ad-hoc distributed system architectures that have evolved over a period of time constrained by the serial and closed nature of the atomic computing unit, the Turing machine discussed above. As von Neumann pointed out in his Lectures given in the Hixon symposium “the basic principle of dealing with malfunctions in nature is to make their effect as unimportant as possible and to apply correctives, if they are necessary at all, at leisure. In our dealings with artificial automata, on the other hand, we require an immediate diagnosis. Therefore, we are trying to arrange the automata in such a manner that errors will become as conspicuous as possible, and intervention and correction follow immediately.

Comparing the computing machines and living organisms, he points out that the computing machines are not as fault tolerant as the living organisms.  He goes on to say “It’s very likely that on the basis of philosophy that every error has to be caught, explained, and corrected, a system of the complexity of the living organism would not run for a millisecond.” The lack of architectural resiliency of cellular organisms in the IT infrastructure reflects itself in the complexity that contributes to 70% of the total IT budget in an organization going into self-maintenance leaving little room for new services development and deployment. While automation of service management has attempted to reduce labor costs, it is often replaced by expensive software, shelf-ware and very expensive knowledge professionals devoted to keep the systems operational by isolating, diagnosing and fixing problems when they occur.

Figure 2: A computing workflow and operating system based management workflow instructions are executed serially by a Turing machine while other management workflows control services that span across multiple physical devices connected by various networks.

Statistics dictate that the frequency of problems becoming visible in a data center grows higher as the number of components grow to meet the ever-increasing demand for communication, collaboration and commerce at the speed of light on a global scale. It is estimated that one company Google alone will require managing 10 million servers in the future to meet its business needs. End to end transaction resiliency (FCAPS) management in such cases becomes extremely critical when service transaction times, service failure times and service repair times are of the same order of magnitude.

Figure 3 shows the world-wide data center spending breakdown presented by IDC in the Server Design Summit (www.ServerDesignSummit.com ). The breakdown of costs show power & cooling costs, management costs and server costs.

The key point to note is that, while server costs are going down with accompanying performance increase, the management costs now dominate the data center expenses. While power and cooling expenses are also increasing (some estimates put it at up to 30% of total data center operational expense), it is now being addressed by redesigning the data centers and migrating to new generation of many-core processors and servers that are designed to drastically reduce power consumption. That leaves the management costs as the next major productivity improvement opportunity. There are two components to this management cost:

1. Hardware infrastructure resiliency (operation & management) assurance cost and
2. End-to-end service transaction resiliency spanning across multiple devices (operation & management) assurance cost.

Figure 3: Breakdown of Costs in a datacenter

While the hardware infrastructure (servers, routers and storage devices) reliability and manageability have drastically improved over the past few decades, a new degree of complexity was introduced to address the reliability of distributed transactions that span across multiple hardware devices. Clustering, server-, storage-, and network- resource provisioning, monitoring and optimization appliances and software systems managing application reliability, availability, performance, security, disaster recovery etc., have all contributed the drastic increase in services management cost. The root cause of the resulting complexity is the ad-hoc nature of distributed systems management and end-to-end transaction service resiliency assurance. As mentioned earlier, the operating systems in individual servers (or virtual servers) do not have visibility and control of end to end transaction resilience and their role of resource management to assure service resilience have been usurped by myriad ad-hoc layers of management. There is no operating system that provides end-to-end resource management of distributed resources. In a distributed system, the computation workflow utilizing the resources and the resource management workflow providing the resources interact with each other to allow for implementing end-to-end visibility and control to meet changing business priorities, workload fluctuations and latency constraints. It is as if an end-to-end distributed resource manager (the observer) is observing the dynamic service workflows and controlling the resources to accomplish the overall service workflow (the observed) goals.

There are four key factors contributing to the lack of resiliency (or contributing to the high cost of resiliency) in distributed systems:

1. When multiple computations compete for resources, management that resolves contention for resources based on global policies external to the individual computations is essential. Operating systems provide this function by executing workflows that manage the local resources and allocate them to appropriate services based on external policies. While the operating systems provide this function implementing both signaling and FCAPS management services in a physical device executing the computations, it cannot control resources across multiple devices.
2. A lack of rigorous computing models to define a composition scheme that defines a distributed computing unit made up of individual computing units (defining a larger “self” or the observer/manager consisting of a group of distributed resources under its control) to appropriately define management hierarchies to implement managed visibility, reflection, and control based on external policies.
3. A lack of dynamic representation of both the observer/manager and the observed/managed and their interactions in a distributed system which has led to ad-hoc implementations contributing to the complexity (of orchestrators orchestrating managers managing agents implementing workflows using TMs) and
4. The propensity of system vendors to sell maintenance intensive products to increase their profit margins through professional services and more software to maintain their products. For example, a particular large enterprise software vendor’s annual maintenance revenue is orders of magnitude higher than their new product revenue.

As the cost of maintenance grows, the search is on, for orders of magnitude improvement with a paradigm shift. The advent of many-core servers with parallelism at the core, performance that scales with the number of cores in a processor, and the hierarchy of high bandwidth networks mentioned earlier provides a new opportunity to look at computing models, distributed systems and their management. One such attempt is the Distributed Intelligent Managed Element (DIME) computing model. Derived from the observations of resilient, efficient and highly scalable distributed systems such as cellular organisms and human network organization, the DIME computing model exploits both the parallelism and high bandwidth to define a new distributed system architecture that integrates the computing workflows and the management workflows by introducing the management of a TM and a parallel signaling network overlay for FCAPS management of a group of managed TMs. The DIME computing model is based on the observation that all computations and workflows can be specified as a directed acyclic graph (DAG). Vertices in a DAG often have a natural ordering – for example, vertices may represent events ordered in time or ordered by hierarchy. The ordering makes the results and algorithms for DAGs relatively simple. DIME network computing executes managed directed acyclic graphs using a set of managed von Neumann Turing machines organized as a network of computing elements with an overlay of signaling enabled control workflow. The DIME computing model allows the specification and execution of a recursive composition model where each computing unit at any level specifies and controls the execution of the DAG at the lower level. The hierarchical composition of “self” representing the manager and the dynamic representation of the observer and the observed allows precise specification of the managed DAG and avoids the thorny problem of halting associated with Turing machines.  Perhaps, DIME networks parallel embriology in biological systems unlike the genetic computing models which embrace mutation. Only theorists can tell.

Theoretically possible or not, or whether it is blessed by the academicians and venture capitalists or not, Figure 4 shows an implementation of DIME computing model that allows the encapsulation of a Linux process as an FCAPS managed DIME and a network of DIMEs providing LAMP (Linux, Apache, MySQL and PHP/perl/Python) based web services.

The prototype was implemented without using Hypervisors, VM infrastructure and a plethora of interfaces to various management systems. Each DIME uses parallel threads to control the computing unit called Managed Intelligent Computing Element (MICE). Each DIME supports a signaling overlay allowing monitoring and control of each DIME’s FCAPS parameters in a distributed DIME network. A novel feature of the DIME network architecture is the decoupling of services management in implementing workflows from the underlying hardware infrastructure management. The services monitor and manage their own response times and other FCAPS parameters using local operating system and if these parameters fall outside the watermarks set by system-wide policies, the workflow is reconfigured. This may involve moving the affected service components appropriately to where the required infrastructure services are available. This allows the infrastructure providers and service providers to negotiate for appropriate service levels a-priori and establish policies on how to manage fluctuations.

Figure 4: The DIME network architecture implementing LAMP services with auto-scaling, self-repair and live-migration without using VM infrastructure.

In addition, the same DIME computing model is also implemented in bare-metal, using a native operating system called Parallax. The Parallax OS converts each core in a many-core processor into a self-managed computing element (DIME) with signaling capability and a signaling network overlay over a computational service delivery network allows an orchestrator to develop, deploy and manage distributed service workflows. A video shows auto-scaling, self-repair, state-ful live-migration and end to end transaction FCAPS management implemented in multi-core servers. (http://youtu.be/K0AxJPaA_RI). This approach seems to suggest that it is possible to use current generation OSs and development environments to create service workflow executable that can be run under the new OS with run time FCAPS management programmed to endow self-resilience.

Whether these are good answers for eliminating the complexity of VM management by eliminating them or not, the DIME network architecture provides an existence proof of an alternative to current complexity. It also points to a direction away from current ad-hoc hacking implementations of distributed systems to where more productivity improvements are possible by searching for computation models that are firmly based on solid theoretical foundations (as Goldin and Wegner suggest) and exploiting the parallelism and high bandwidth now available in hardware. Kuhnian paradigm shift or not, hopefully there are alternatives that will change the direction of IT from current “Buy now and pay-forever for maintenance software and services” philosophy which is contributing to the cost and complexity. Hopefully, this will also eliminate our current strategy to throw human resources and more software to diagnose and fix problems than building self-correcting systems as cellular organisms do. As von Neumann was seeking,  the decoupling of hardware infrastructure management from services management may point to a direction where building reliable services using not so reliable hardware is possible as the cellular organisms do.

If you have transformational research results, or want to make a real difference in computer science research, see Call for Papers at:

www.workshop.kawaobjects.com and http://WETICE.org

This emphasis on the sensory monitoring of the environment, dynamic coupling, connectivity and system-wide coordination is also confirmed by observations on cell communication.  According to biologist Sean B. Carroll [2], “cells communicate with one another by sending signals in the form of proteins that are exported and travel away from their source.  Those proteins then bind to receptors on other cells, where they trigger a cascade of events, including changes in cell shape, migration, the beginning or cessation of cell multiplication, and the activation or repression of genes.”

Cellular organisms developed very sophisticated computing models well before their brain evolved. The architectural resiliency of cellular organisms stems from their ability to manage highly temporal phenomena. System-wide connectivity and coordination require a sense of time, history and synchronization between various tasks performed by a group of loosely coupled elements which, as Louis Barrett points out, the Turing machine implemented using the stored program control lacks.  Discussing the nature of temporal phenomena, she writes “This means simply that the actual rates and rhythms that characterize a particular process play an important and central role in getting the job done.  This could be the way that the underlying physical processes of the brain work (how long it takes for a neurotransmitter, like nitric oxide or glutamate, to diffuse through the brain, for example, or how long it takes for such neurotransmitters to modulate neuronal activity), which in turn could affect the specific duration or rates of change in other physiological processes.  Similar intrinsic rhythms in the body may also be important, as will other aspects of the body dynamics that relate to, for example, the mechanical properties of the muscle, which dictate where and how fast an animal can move.  These bodily processes may, in turn, need to be synchronized precisely with temporal processes occurring outside of the animal in the environment.”  She also points out that the coordination and synchronization requires system-wide information processing and routing that the brain provides.

Compare this with the quest for real-time information processing currently being driven by global communication, collaboration and commerce at the speed of light.  Whether it is high frequency trading, web-based commerce, social networking or federated enterprise computing, the ability to manage highly temporal phenomena in real-time is becoming critical. System-wide connectivity, high availability, security and performance management require coordination with a sense of time, history and synchronization between various tasks performed by a group of loosely coupled elements.  There are two drivers behind the search for new computing models that go beyond current von-Neumann computing model:

1. Poor end-to-end distributed transaction reliability, availability, performance and security as recent episodes at Sony, Amazon, Google, and RSA [3, 4, 5 and 6] demonstrate.
2. The hardware upheaval caused by the new class of many-core processors that allow parallelism which cannot be fully exploited with current state of software innovation.

The DIME network architecture introduced in WETICE 2010 [7] exploits parallelism and addresses the end-to-end distributed transaction management using a signaling network overlay over a network of von Neumann stored program control (SPC) computing nodes to implement dynamic fault, configuration, accounting, performance, and security management of both the nodes and the network based on business priorities, workload variations and latency constraints.  The following video explains the differences between the DIME networks implementing a network of managed Turing machines and the von-Neumann computing architecture.

Figure 1 shows the screenshot of DIME implementation in Linux Operating System (OS)

DIME network architecture in Linux Operating System

Two implementations of DIME networks demonstrate the feasibility of this architecture [8, 9].  In one implementation, a Linux process is encapsulated as a DIME to create a DIME network.  Figure 1 shows a screenshot of a 6-DIME network executing a simple c program in parallel.  Each DIME is programmed to self-manage by monitoring heartbeat, performance etc.  A supervisor DIME implements recovery policies and end-to-end workflow management.  In the second implementation, a native OS called Parallax is implemented from scratch in assembler with c and c++ interface to encapsulate each core as a DIME.  The following video shows the self-repair feature of DIME network architecture implemented using parallax Operating system in a multi-core server network.

As multiple reviewers of the DIME papers noted, this approach is “novel and interesting”.  It attempts to bring the architectural resiliency of cellular organisms by exploiting the parallelism offered by the multi-core architecture and the signaling abstractions that are essential for implementing dynamic temporal systems.  However, as pointed out in one of the papers [8], “The history of the evolution of current OSs is filled with lessons on wasted billions (does anyone remember Multics or OS2?), unmet expectations (who would have thought UNIX, the original System V, would vanish), surprise winners (Windows and Linux), and stealthy survivors (Mach in a Mac).”

Figure 2 shows the DIME network architecture compared with conventional computing and cloud/grid computing architectures in terms of resiliency, efficiency and scaling.  The resiliency is measured with respect to a service’s tolerance to faults, fluctuations in contention for resources, performance fluctuations, security threats and changing business priorities.  Efficiency is measured in terms of total cost of ownership and return on investment.  Scaling addresses end-to-end resource provisioning and management with respect to increasing number of computing elements required to meet service needs. Current operating systems do not easily scale because of their dependence on current lock mechanisms, thread technology implementations and their size and complexity.

Figure 2: Resiliency, Efficiency and Scaling of Distributed Systems

As the picture depicts, the grid and cloud computing paradigms automate many of the administrative tasks in assuring service management while the DIME (Distributed Intelligent Managed Element) network architecture attempts to provide dynamic self-management of service workflows.  The cloud and grid architectures do not scale easily and increase complexity with layers of management systems to compensate for the serial von Neumann implementation of the service nodes that do not address temporal dynamics of distributed systems.

The DIME network architecture is a departure from conventional wisdom currently being pursued by the universities and corporate R&D.  It adds monitoring and control of each Turing computing node and a signaling enabled network to implement the management of temporal behavior of workflows executed as directed acyclic graphs using a network of managed Turing machines.  The concept of a parallel signaling channel is foreign to the current generation of IT professionals, except for those with telecommunications or voice over IP experience. (Unfortunately, the dismantling of research organizations such as AT&T Bell Labs which were the guardians of institutionalized technical knowledge, increasing commercialization of university research and Goldman-Sachs-like investment flipping philosophy to pursue instantaneous profits have put a dent on disruptive innovation and preservation of institutionalized know-how.) Signaling allows establishing equilibrium patterns and monitor and control exceptions system-wide.  It allows contention resolution based on system-wide view and eliminates race conditions and other common issues found in current distributed computing practice. In systems with strong dynamic coupling between various elements of the system, where each change in one element continually influences other element’s direction of change, signaling in the computational model helps implement system-wide coordination and control based on system-wide priorities, workload fluctuations and latency constraints. The DIME network architecture is either profoundly disruptive (disruptive to the vendors of high maintenance software, shelf-ware and recurring services but not to their customers who will benefit from the simplicity of services management with temporal dynamics at both the node and network level)  or it is like many other ideas that will go nowhere for many odd reasons.  The history of IT is filled with many forgotten or mismanaged innovations  [10, 11 and 12] .  It is hard to predict which solution, evolution prefers in its continuous quest for lower entropy.  Some times it favors revolutions and at other times it is satisfied with incremental solutions.

But again, as Mitchell Waldrop points out, revolutions are not revolutions if they are believed in at the start. Are they?

“How did it go in Berkeley? Did they like your ideas?”
“It was the pits,” said Arthur. “Nobody there believes in increasing returns.”
Susan Arthur had seen her husband returning from the academic wars before. “Well,” she said, trying to find something comforting to say, “I guess it wouldn’t be a revolution, would it, if everybody believed in it at the start?”

—        Waldrop, M.M., “complexity: The Emerging Science at the Edge of Order and Chaos”, New York, Simon and Schuster, (1992) p 19.

References:

1. Barrett, L., “Beyond the Brain: How Body and Environment Shape Animal and Human Minds,” Princeton University Press, Princeton, 2011, p 116, 122
2. Carroll, S. B., “The New Science of Evo Devo - Endless Forms Most Beautiful”, New York: W. W. Norton & Co. 2005, p74.
3. Morris, C. (2011). Sony PlayStation Facing Yet Another Security Breach, New York, CNBC.com (http://www.cnbc.com/id/43079509 )
4. Thibodeau, P. and Vijayan, J. (2011). Amazon EC2 service outage reinforces cloud doubts. Computerworld (http://www.computerworld.com/s/article/356212/Amazon_Service_Outage_Reinforces_Cloud_Doubts)
5. http://www.searchenginejournal.com/googles-downtime-affected-5-of-the-internet/10463/
6. Moscaritolo, A. “RSA confirms Lockheed hack linked to SecurID breach,” 2011, SC MAGAZINE, June 07
   (http://www.scmagazineus.com/rsa-confirms-lockheed-hack-linked-to-securid-breach/article/204744/ )
7. Rao Mikkilineni and Giovanni Morana, “Is the Network-centric Computing Paradigm for  Multi-core, the Next Big Thing?” ( http://computingclouds.wordpress.com )
8. Morana, G., and Mikkilineni, R., “Scaling and Self-repair of Linux Based Applications Using a Novel Distributed Computing Model Exploiting Parallelism”. IEEE proceedings, WETICE2011, Paris, 2011
9. Mikkilineni R., and Seyler, I. “Parallax – A New Operating System Prototype Demonstrating Service Scaling and Self-Repair in Multi-core Servers”, IEEE proceedings, WETICE2011, Paris, 2011
10. Can Cisco Sustain Competitive Differentiation on Operational Excellence Alone? (www.metooeconomist.wordpress.com)
11. Acquisitions, Innovation and the Economics of the Invisible Hand (www.metooeconomist.wordpress.com)
12. Are the Short-Term Profit Motives and Wall Street-like Investing under the Influence, Trumping Long Term Innovation in the Silicon Valley?

Dr. Rao Mikkilineni

IEEE Member

Kawa Objects Inc.,

Los Altos, California, USA

rao@kawaobjects.com

and

Dr. Giovanni Morana

DIEEI

University of Catania

Catania, Italy

giovanni.morana@dieei.unict.it

The stated objective of the first workshop on Collaboration and Cloud Computing” in WETICE 2009 was “to analyze current trends in Cloud Computing and identify long-term research themes and facilitate collaboration in future research in the field that will ultimately enable global advancements in the field that are not dictated or driven by the prototypical short-term profit driven motives of a particular corporate entity.”

This track discusses the progress made within the span of three conferences which helped develop a new approach to the convergence of distributed clouds, grids and their management. This track presents a new computing model and its implementation which resulted directly from the collaboration of the two workshops, Collaboration and Cloud Computing Workshop and the Emerging Technologies for Next-Generation grids sponsored under the Aegis of WETICE. In this track we discuss one of the key issues that still need to be addressed to improve the efficiencies and utilize the new generation of many-core servers that are transforming the information technology landscape.

Current Information Technology solutions have become silos of server, storage and network infrastructure with poor end-to-end distributed transaction reliability, availability, performance and security as recent episodes at Sony, Amazon, Google, and RSA [1, 2, 3 and 4] demonstrate.  We believe that there is a need for reexamining the fundamental architectural foundation of Information Technologies to transform the data centers from their current role of being just managed server, networking, and storage hosting centers (whether physical or virtual), to true service switching centers with telecom grade trust.  We need a paradigm shift from resource switching and connection management to services switching and service connection management.  We also believe that new approaches are essential to replace the current efforts to replicate the complexity inside the data center today, also inside the many-core servers.  We hope WETICE2011 will continue the tradition to forge new collaborations that lead to innovation without a short term agenda based on immediate profit motives proposing incremental solutions which do not address fundamental cost and complexity issues. An opportunity exists to exploit the hardware upheaval unleashed by the many-core chips with equally innovative software solutions.

We conclude by quoting von Neumann [5] ”The basic principle of dealing with malfunctions in nature is to make their effect as unimportant as possible and to apply correctives, if they are necessary at all, at leisure. In our dealings with artificial automata, on the other hand, we require an immediate diagnosis. Therefore, we are trying to arrange the automata in such a manner that errors will become as conspicuous as possible, and intervention and correction follow immediately.” Comparing the computing machines and living organisms, he points out that the computing machines are not as fault tolerant as the living organisms. He goes on to say “It’s very likely that on the basis of philosophy that every error has to be caught, explained, and corrected, a system of the complexity of the living organism would not run for a millisecond.”

—  Neumann, J. v., General and Logical Theory of Automata. edited and compiled by William Aspray and Arthur Burks, MIT Press, 1987, p408

15 Papers received: 6 rejected, 6 accepted as FULL papers and 3 accepted as SHORT papers

For the Agenda, the Venue and accommodation, please visit

http://events.telecom-sudparis.eu/wetice/program/index_cdcgm.php

References:

1. Morris, C. (2011). Sony PlayStation Facing Yet Another Security Breach, New York, CNBC.com (http://www.cnbc.com/id/43079509 )
2. Thibodeau, P. and Vijayan, J. (2011). Amazon EC2 service outage reinforces cloud doubts. Computerworld (http://www.computerworld.com/s/article/356212/Amazon_Service_Outage_Reinforces_Cloud_Doubts )
3. http://www.searchenginejournal.com/googles-downtime-affected-5-of-the-internet/10463/
4. Moscaritolo, A. RSA confirms Lockheed hack linked to SecurID breach, 2011, SC MAGAZINE, June 07 (http://www.scmagazineus.com/rsa-confirms-lockheed-hack-linked-to-securid-breach/article/204744/ )
5. Neumann, J. v. (1987). General and Logical Theory of Automata. edited and compiled by William Aspray and Arthur Burks, MIT Press, p408

Co-Chairs

  1st Track on Convergence of Distributed Clouds, Grids and their Management;
A combined track of the 3rd CCC and the 8th ETNGRID

 WETICE 2011 – Conference: 20^th IEEE International Conference on Collaboration Techniques and Infrastructure 

Summary:

WETICE 2011 track on Convergence of Distributed Clouds, Grids and their Management to be held in Paris (June 27 – 29) has already received some interesting papers.  They include papers on new computing models, Clouds, Grids, their management and also some proofs of concept demonstrations.  The Proceedings will be published by the IEEE CS Press and distributed at the conference.  All published papers are refereed.  The last date for submission of papers is March 5, 2011.

This blog gives a preview of the ideas that will be discussed in June to attract other potential papers that could make a major impact on next generation clouds, grids and their management.

Master Foo and the Old Hand [1]

An experienced UNIX programmer, hearing of Master Foo’s wisdom, came to him for guidance. Approaching the Master, he bowed three times and said:

“Master Foo, I am gravely troubled. In my youth, those who followed the Great Way of Unix used software that was simple and unaffected, like ed and mailx. Today, they use vim and mutt. Tomorrow I fear they will use KMail and Evolution, and Unix will have become like Windows — bloated and covered over with GUIs.”

Master Foo said: “But what software do you use when you want to draw a poster?”

The programmer replied: “I…have never done that. But I am sure that I could use LaTeX or pic to accomplish it without GUIs, in the proper UNIX way.”

Master Foo then said: “Which one will reach the other side of the river: The one who dreams of a raft, or the one that hitchhikes to the next bridge?”

Upon hearing this, the programmer was enlightened.

From POTS, PANS, and SANs to Clouds, Grids and Their Management:

Ma Bell (Known as AT&T) in 1907 introduced the standard for a voice service in a cloud when Theodore Vail, the then president of AT&T, unveiled a vision for providing universal service and telecom grade trust (providing reliable, secure and high performance connection at a reasonable cost).  The service was analog.  The operation was manual.  The dial tone, introduced to assure the telephone user that the exchange is functioning when the telephone is taken off-hook by breaking the silence (before an operator responded) with an audible tone, has become a symbol for universal service and telecom grade trust.  Later on, the automated exchanges provided a benchmark for telecom grade trust that assures managed resources on-demand with high availability, performance and security.  Today, as soon as the user goes on hook, an intelligent network recognizes the user profile based on the dialing telephone number.  As soon as the destination party number is dialed, the network recognizes the destination profile and provisions all the network resources required to make a secure connection authenticating services usage, commences billing, monitors and assures the connection till one of the parties initiates a disconnect.  The history, from the days of (upper case) AT&T’s inception to the days when it was transformed to (lower case) at&t  in 2005, is filled with lessons on  regulation, brilliant Bell Labs innovation, the rise, missed or mismanaged business opportunities, deregulation, unscrupulous competitors such as WorldCom who cooked the books, the fall and the rebuilding of a great corporation in the United States.

A century later, the computing industry is rediscovering the same lessons in offering cloud based computing services.  Virtual computing services offered in the cloud are transforming the way consumers and businesses communicate, collaborate and conduct commerce at the speed of light.  The infrastructure makers, the service developers and the service operators are striving to capture the big chunk of a large market share by racing to provide universal access to virtual computing services with telecom grade trust.  There is a battle brewing between the “Appliances for High Performance Camp” and the “Open System Software and Services Approach for Everything Camp” of the current IT infrastructure vendors.  Figure 1 presents a reference model that shows the various stakeholders and their stakes in the current day cloud computing market.  The products and services that have evolved bottom up in the services stack from the server, network, storage or application and infrastructure software domains are expanding their reach into other domains to gain market share.

Current Cloud Market Landscape

Today, computing virtualization is provided with Hypervisor technology to create virtual servers, network virtualization is provided through multi-protocol routers and switches and storage virtualization is provided through specialized appliances supporting NAS and SAN.  New appliances are being rolled out for databases and storage transaction management.  Different virtualization platforms and orchestrators that integrate them are flooding the market.  The costs of associated services are skyrocketing.  Figure 1 shows various layers of management to provide application specific availability, reliability, performance, security and billing functions.  Various vendors play at various levels in each layer.  The complexity –  of heterogeneity, multiple vendor solutions and orchestrators that provide integration –  has been overwhelming the service developers, operators and consumers.

The situation is very similar to the days before Strowger’s switch eliminated many operators sitting in long rows plugging countless jacks into countless plugs and reducing the cost of adding new subscribers that had risen in a geometric proportion.  It was estimated that the management cost of a new subscriber was more than the revenue that the subscriber contributes.  According to the Bell System chronicles, one large city general manager of a telephone company at that time wrote that he could see the day coming soon when he would go broke merely by adding a few more subscribers [2].  The only difference between today’s IT data center and central office before Strowger’s switch is that “fewer – but very expensive consultants, countless hardware appliances, and countless software systems that manage them” replace “many operators, countless plugs and countless jacks”.  In addition, we have to account for the shelf-ware, the latency introduced by today’s systems administration paradigm (albeit automation ala RightScale and Sclr approach) and costs involved through the services business that has come to dominate the IT vendor revenue streams to help manage the complexity (which was sold in the guise of improving productivity and lowering Total Cost of Ownership).  It is estimated that 60% to 70% of IT data center cost is in its operation and management with or without virtualization in spite of a 10X improvement in hardware, space and energy savings with the new class of servers available today [3].  Figure 2 shows percentage Total Cost of Ownership (TCO) (for a 1500 server data center) over five years by each component with and without virtualization.

Five Year TCO of Virtualization According to a Vendor ROI Calclulator

While virtualization introduces many benefits such as consolidation, real-time business continuity and elastic scaling of resources to meet wildly fluctuating workloads, it adds another layer of management systems in addition to current computing, network, storage and application management systems.  Figure 3 shows a reduction by 50% of the five-year TCO with virtualization.  The Virtual Machine density of about 13 allows a great saving in hardware costs which is somewhat off-set by the new software, training and services costs of virtualization.

TCO over 5 Years with virtualization of 1500 servers using 13 VMs per Server

In addition, there is the cost of new complexity in optimizing the 13 VMs within each server in order to match the resources (network bandwidth, storage capacity, IOPs and throughput) to application workload characteristics, business priorities and latency constraints.  According to a storage consultant, Jon Toigo [4] “Consumers need to drive vendors to deliver what they really need, and not what the vendors want to sell them. They need to break with the old ways of architecting storage infrastructure and of purchasing the wrong gear to store their bits: Deploying a “SAN” populated with lots of stovepipe arrays and fabric switches that deliver less than 15% of optimal efficiency per port is a waste of money that bodes ill for companies in the areas of compliance, continuity, and green IT.”

Figure 4 shows the transition from managed physical server farms to managed virtual server farms:

Transition From Managed Server Farms to Managed Virtual Server Farms

The cost per VM is estimated to be around $2500 with minor variation with the use of VMWare, Microsoft, Red Hat or Citrix solution.  This is consistent with 2X improvement with a managed physical server cost of about $5000.  In spite of vendor claims, there is not much difference between different Hypervisors just as there was no big difference between DB2, Oracle and Sybase in the 1980′s.  They are all equally knowledge intensive and require expensive services solutions to maintain.  Further improvements in TCO have to come from new approaches that drastically reduce complexity of current layers of management systems.  These approaches may have to leverage hardware assisted management functionality in computing infrastructure and new software technologies that support self-configuring, self-monitoring, self-securing, self-healing and self-optimizing architectures. Perhaps it is time for designing a new class of autonomic systems that satisfy their eight defining characteristics [5].  It suffices to say that none of the systems in our data centers today come close to satisfying these defining characteristics.  They are worth looking up!

With huge maintenance and service revenues at stake, current IT vendors do not have incentives to invest in R&D that develops autonomic systems [6, 7 and 8] which will destroy their recurring revenue model.  One IT vendor makes four to five times more revenue annually from accumulated maintenance and services contracts than from new product sales with a captive customer base.  It seems that it does pay to design high maintenance systems.   Without questioning, we pay for security software that fixes the problems with operating systems that should not be there in the first place.  History again provides clues to this behavior.  Ma Bell was very reluctant to deploy fiber because it cannibalized its revenues from investments in copper.  It took deregulation to unleash the fiber. AT&T was dragged kicking and screaming to digital switching by the introduction of DMS switches by competing Nortel.   If history is any guide, perhaps this is an opportunity for new players to eat current IT player’s lunch, breakfast and dinner with fresh ideas and new R&D. Or perhaps we have to wait for a Dragon Warrior [9].

Dragon warrior or not, fortunately, high bandwidth wire-line and wireless networks, multi-core multi-CPU servers with hardware assisted virtualization and management features incorporated at the chip level are offering new alternatives which potentially can simplify the IT computing architecture and reduce the operation and management costs further.  The next conference on “Convergence of Distributed Clouds, Grids and their Management” sponsored under the Aegis of WETICE 2011, to be held at “Institut Telecom, Telecom Sud Paris”, Paris, France (www.telecom-sudparis.eu) from June 27th to 29th, 2011, is devoted to discuss new approaches that go beyond current state of the art.  We are inviting bold and new ideas from corporate R&D (if there still exists Bell Labs class innovation in fundamental computer science to create next generation von Neumann or Masater Foo), universities and entrepreneurs to propose and demonstrate potential next generation systems and ideas without limits on imagination.  What sets WETICE apart from larger conferences is that the conference tracks are kept small enough to promote fruitful discussions on the latest technology developments, directions, problems, and requirements. Each track includes paper presentations and group discussions while the keynote sessions and summary of discussions take place in joint sessions. WETICE welcomes papers on “work-in-progress” from Ph.D. students whose imagination is not often limited by commercial feasibility or financial agendas [7 and 8].

We are already receiving some interesting papers.  For example, an implementation of  a new Distributed Intelligent Managed Element (DIME) Network computing model [3] proof of concept will be demonstrated.  DIME networks could provide a network computing model to create distributed computing clouds and execute distributed managed workflows with high degree of agility, and managed reliability, availability, performance, security and utilization of distributed computing, storage and network resources.  While this computing model can be implemented using current generation IT infrastructure, this paradigm is ideally suited to utilize fully the new generation of multicore multi CPU servers to implement highly scalable, parallel and distributed systems transcending physical infrastructure or geographical boundaries.

Three different implementations of the DIME computing model will be demonstrated:

1. Using a bare metal OS called Parallax which is built ground up to leverage parallelism and multicore chips in 64 bit architectures, the DIME network is implemented to execute managed workflows.  This implementation alleviates the need for a Hypervisor based virtual server and allows distributed managed computing services to be delivered with universal access and telecom grade reliability.
2. By implementing the DIME networks using Virtual Servers loaded with just enough Linux OS compatible with current cloud architectures
3. By implementing a just enough Linux OS to create DIME networks on  current generation  servers

Figure 5 shows the parallax version of DIME network implementation that eliminates the need for a Hypervisor and shows the service creation, delivery and assurance platform architecture.

DIME Network Computing Model and Distributed Parallel Computing Architecture

The DIME network computing model proposes a signaling network overlay over the computing network and allows parallelism in resource monitoring, analysis and reconfiguration based on workload variations, business priorities and latency constraints of the distributed software components.  A workflow is implemented as a set of tasks, arranged or organized in a directed acyclic graph (DAG) and executed by a managed network of distributed computing elements (called Distributed Intelligent Managed Elements, DIMEs). These tasks, depending on user requirements are programmed and executed as loadable modules in each DIME.  Figure 6 shows the DIME network based services creation, deployment and assurance framework.

Potential Distributed Services Creation, Delivery and Assurance Framework

In addition, WETICE 2011 will present papers on the state of the art and future trends in understanding the convergence of clouds, grids and their management.

We are seeking more new ideas to discuss in the conference.

The goals of the conference include (but are not limited to):

1. Discovering new application scenarios, proposing new operating systems, programming abstractions and tools
2.  Identifying the challenging problem that still need to be solved such as parallel programming, scaling and management of distributed computing elements, and
3.  Reporting results and experiences gained by researchers in building dynamic Grid-based middleware, computing clouds (distributed or otherwise) and workflow management systems.

Various deadlines to participate in the conference are as follows:

• Paper submissions deadline:                              March 5, 2011
• Decision to paper authors:                                  April 4, 2011
• Camera Ready papers to IEEE:                           April 29, 2011
• WETICE-2011 conference:                                  June 27-29, 2011

All papers submitted will be reviewed by peers and selected papers will be published in WETICE2011 Conference Proceedings by IEEE.

All papers must be submitted to workshop@kawaobjects.com,

Details of the conference and past conference archives can be accessed at www.workshop.kawaobjects.com,

http://etngrid.diit.unict.it and

http://wetice.org

References:

1. http://catb.org/esr/writings/unixkoans/oldhand.html
2. http://www.telephonetribute.com/switches.html#Mr. Almon B. Strowger and His Electric Telephone Switch
3. Rao Mikkilineni and Giovanni Morana, “Is the Network-centric Computing Paradigm for Muti-core, the Next Big Thing?” ( http://computingclouds.wordpress.com )
4. Jon Toigo, http://www.datastorageconnection.com/article.mvc/Jon-Toigo-Exposes-More-About-Data-Storage-Ven-0001
5. http://www.research.ibm.com/autonomic/overview/elements.html
6. Acquisitions, Innovation and the Economics of the Invisible Hand
7. Are the Short-Term Profit Motives and Wall Street-like Investing under the Influence, Trumping Long Term Innovation in the Silicon Valley?
8. Organizational DNA, Disruptive Innovation, Economics at the Speed of Light and the Impact on Investment in Future
9. Kung Fu Panda, The Movie, Dream Works, 2008

 Dr. Rao Mikkilineni 

Chair, Collaboration and Cloud Computing Workshop, WETICE2010    

and    

Dr. Giovanni Morana    

Chair, ETNGRID Workshop, WETICE2010  

 “New source of competitive differentiation may lie in the internal capacity to reconfigure resources in real-time”  C. K. Prahlad and M. S. Krishnan, “The New Age of Innovation” McGraw Hill, NY2008

 Four major technology trends are radically altering the way we use and manage computers to conduct our personal and business affairs: 

1. Multi-core, multi-CPU chips and servers with hardware assisted virtualization that offer 10X performance improvement in space, energy and computing while decoupling services from the hardware on which they are deployed [1],
2. Integration of management at the chip level to provide a higher degree of reliability, availability, performance and security [2],
3.  Highly scalable parrallel computing architectures that can be deployed in mobile. Laptop and enterprise class hardware and 
4.  High bandwidth wire-line and wireless networks that provide high degree of mobility and interactivity 

Resulting services supporting communication, collaboration and commerce at the speed of light transcend not only geographical boundaries but also physical infrastructure  boundaries thus enabling truely distributed computing and execution of workflows limited only by their latency constraints. 

However, the progress in hardware at the chip level is not matched by the progress in our ability to leverage the parrallelism and scaling possible with our softare and computing models available today [3].  In fact, current generation operating systems, programming languages and frameworks are derived from a computing model that has evolved from an environment where the CPU resources were limited and shared by multiple clients to optimally utilize the available resources.  The distributed computing models also were constrained by low bandwidth networks supported by special purpose hardware such as routers, switches and storage network systems.  This has given rise to a layer of server, network, storage and application management systems[1] required to assure reliability, availability, performance and security of business workflows implemented using shared CPU, network and storage resources.  Parrallelism and scaling are constrained by the sequential computing models at the chip level.  

The new multi-core multi-CPU computing architecture allows high bandwidth networking inside the server to exploit the parrallelism.  It is therefore, appropriate to reexamine the current computing models and create next generation distributed workflows that leverage the features offered by hardware assisted virtualization, management at the chip level, parallelism, scaling and high-bandwidth wire-line and wireless network connectivity.   Dynamic real-time reconfiguration of infrastructure resources is made possible with virtualization technologies.  Efficient and dynamic real-time reconfiguration of business processes require new computing and programming models that exploit parallelism and reduce the semantic gap between workflow descriptions at high level and their execution at the chip level.  

Furthermore, hardware assisted virtualization, chip level management and the ubiquitous availability of high bandwidth networks allow overcoming  of today’s cloud service limitations, based on islands of proprietary architectures delivered by mega-providers, and enable the creation of federate distributed clouds [4] composed by small, medium, and large providers that cooperate with each other with telecommunication grade trust which is taken for granted in telephony and the Internet. 

The conference on “Convergence of Distributed Clouds, Grids and their Management” sponsored under the Aegis of WETICE 2011 is devoted to addressing next generation computing models which support real-time resource reconfiguration of distributed business workflow execution based on latency constraints, changing workloads and business priorities.  It is devoted to addressing the assurance of reliability, availability, performance, account management and security of distributed business process execution with appropriate visibility and control.  

The target is to transform current labor and knowledge intensive IT management into self-configuring, self-monitoring, self-healing and self-optimizing distributed workflow implementations with resource management only limited by the speed of light. 

This conference is a result of combining the two workshops the ETNGRID (http://etngrid.diit.unict.it/) and Collaboration and Cloud Computing (www.workshop.kawaobjects.com) conducted in the International IEEE sponsored WETICE2010 (http://WETICE.org/) held in TEI, Larissa, Greece [5]. 

The conference will be held at “Institut Telecom, Telecom Sud Paris”, Paris, France (www.telecom-sudparis.eu) from June 27th to 29th, 2011. For more details on the venue, go to http://wetice.org and download WETICE 2011 presentation (pdf) for more details.

1. Discovering new application scenarios, proposing new operating systems, programming abstractions and tools 

2.  Identifying the challenging problem that still need to be solved such as parallel programming, scaling and managemet of distributed computing elements, and  

3.  Reporting results and experiences gained by researchers in building dynamic Grid-based middleware, computing clouds (distributed or otherwise) and workflow management systems. 

• Paper submissions deadline:                              March 5, 2011 
• Decision to paper authors:                                  April 4, 2011 
• Camera Ready papers to IEEE:                          April 29, 2011 
• WETICE-2011 conference:                                   June 27-29, 2011 

All papers submitted will be reviewed by peers and selected papers will be published in WETICE2011 Conference Proceedings by IEEE. 

All papers must be submitted to workshop@kawaobjects.com, 

Details of the conference and past conference archives can be accessed at www.workshop.kawaobjects.com, 

http://etngrid.diit.unict.it and 

http://wetice.org 

References:

[1] Vijay Sarathy, Purnendu Naraian, Rao Mikkilineni, “Next generation Cloud Computing Architecture – Enabling real-time dynamism for shared distributed physical infrastructure”, Proceedings of IEEE WETICE 2010, Second International Workshop on Collaboration & Cloud Computing, June 2010 (http://www.kawaobjects.com/resources/PID1258479.pdf ) 

[2] Arvind Kumar, Purushottam Goel, and Ylian Saint-Hilaire, “Active Platform Management Demystified”, Intel Press, 2009 

[3] David Patterson, “The trouble with multi-core”, IEEE Spectrum, July 2010, p28   

[4] T. Bittman, “The evolution of the cloud computing market,” Gartner Blog  Network,http://blogs.gartner.com/thomas bittman/2008/11/03/theevolution-of-the-cloud-computing-market/, November 2008. 

[5] Is the Network-centric Computing Paradigm for Muti-core, the Next Big Thing? 

Chair, Collaboration and Cloud Computing Workshop, WETICE2010

and

Dr. Giovanni Morana

Chair, ETNGRID Workshop, WETICE2010

“During the past several decades, the general-purpose microprocessor industry has effectively leveraged Moore’s Law to offer continually increasing single-thread microprocessor performance and compelling new features. However, the amazing increase in performance was not free: many practical design constraints, especially power consumption, were pushed to their limits. The integration of multiple microprocessor cores into CPU chips has improved the capability of the single-CPU chip systems and extended the capability of the multiple-CPU chip systems in a very natural way. General-purpose multi-core processors have brought parallel computing into the mainstream and penetrated markets from laptops to supercomputers.” Chuck Moore and Pat Conway [1]

According to Intel, a product from its multi-core product lineup, in addition to its multi-core design, will have more than 1.7 billion transistors and 24 megabytes of cache memory. Other features Intel provides as part of its chip design capability is the virtualization technology (codenamed Vanderpool Technology or VT), which will allow computers to run multiple operating environments simultaneously and significantly reduce reliability issues. Other vendors such as AMD have joined the fray with competing product lines offering multi-core processors and hardware assisted virtualization.  DELL, HP and other server vendors are touting a 10X performance improvement along with space, power, cooling and cost advantages.  Their product lines span from laptops to enterprise-class servers with highly scalable architectures.

What is the Problem with Multi-core?

More recently, David Patterson [2] asserts that taking real advantage of multi-core processors is limited by our inability to figure out how to make dependable parallel software that works efficiently as the number of cores increases.  He concludes with a pessimistic note. Talking about dependable parallel software, he says “Although I’m rooting for this outcome – and many colleagues and I are working hard to realize it -  I have to admit that this third scenario is probably not the most likely one.”  He cites many casualties from the past such as “Ardent, Convex, Encore, Floating Point Systems, Inmos, Kendall Square Research, MasPar, nCube, Sequent, Tandem, and Thinking Machines” as the most prominent names from a long list of long-gone parallel hopefuls.  He also cites the list of languages designed to support parallel processing and points out their inadequacy in making parallel programming easy or straight forward.

Is the Server-centric OS in a Multi-core, a Square Peg in a Round Hole?

While the evolution of current software architectures starting from the operating system have evolved from a server centric architecture where the CPU resource is scarce and is shared to perform multiple tasks, the multi-core Processors promote a contra architecture where each CPU has multiple threads and multiple CPUs can be networked to perform tasks in parallel.

As demonstrated by nature, the ability to work in parallel as a group represents a very efficient way to reach a common target.  In fact, through evolution over a long period of time, the insects, the animals and, mainly, the human beings have learned to aggregate themselves in a similar architecture where each entity/node may have modest (computational) ability but a network of entities/nodes can organize themselves to accomplish goals that an individual entity/node itself cannot.

The human networks are considered intelligent because they accomplish their goals in multiple ways by collecting information from the external world and using it to control their environment.  The human network consists of a group of individuals (Computing entities) operating as a system [3]:

1. Every system has a purpose within a larger system
2. All of a system’s parts must be present for the system to carry out its purpose optimallyA system’s parts must be arranged in a specific way for the system to carry out its purpose (separation of concerns)
3. Systems change in response to feedback (collect information, analyze information and control environment using specialized resources)
4. Systems maintain their stability (in accomplishing their purpose) by making adjustments based on feedback

According to Vancho Cirovski, [4], the effectiveness of the human network depends on the connections, communication and mastery (or specialization) of the individual human object.  Better the quality of mastery of the individual node, the quality of connection and communication, higher the effectiveness.

Humans have created organizational frameworks through evolution.  According to Malone [5], organization consists of connected “agents” accomplishing results that are better than if they were not connected.

An organization establishes goals, segments the goals into separate activities to be performed by different agents, and connect different agents and activities to accomplish the overall goals.  Scalability is accomplished through hierarchical segmentation of activities and specialization.

There is always a balance between the cost of coordination of the agents and economies of scale obtained from increasing the network size.  As the size of the group increases, efficiency of the organization is achieved through specialization and segmentation.  On the other hand agility of an organization depends on how fast the organization can respond to changes required to accomplish the goals by reconfiguring the network. This is accomplished using a communication mechanism that allows each agent to exchange information with others.

The specialization and segmentation characteristics of human networks, the distributed coordination and the ability to dynamically adapt to changing scenarios, represent an effective schema that, if emulated in the multi-core CPUs networks, can be useful to improve their behavior, both in term of computational performance and in term of robustness, availability, reliability, performance and security. An important step, in this direction, consists in the introduction of a FCAPS-aware management framework.

A framework able to address Fault, Configuration, Accounting (utilization), Performance and Security of all network elements (in this case the agents) is the key to guarantee both the efficiency and the agility seen in the human networks.

The Telecommunications Management Network (TMN) defines the FCAPS framework:

1. Fault management, by detecting and correlating faults in network devices, isolating faults and initiating recovery actions
2. Configuration management, by providing change tracking, configuration, installation and distribution of software to all network devices
3. Accounting management capability through comprehensive network usage reports generated by collecting and parsing accounting data
4. Performance management by providing real-time access for the monitoring of network performance (QoS) and resource allocation data
5. Security management by providing granular access control for network resources

Project management is a specific example where Fault, configuration, accounting, performance and security are individually managed to provide an optimal network configuration with a coordinated work-flow.  Functional organizations, and hierarchical and matrix organizational structures are all designed to improve the efficiency and agility of an organization to accomplish the goals using both FCAPS management and signaling.

Connection management is achieved through effective communications framework.  Over time, human networks have evolved various communications schemes and signaling forms the fundamental framework to configure and reconfigure networks to provide the agility.  There are four basic abstractions that comprise signaling:

1. Alerting,
2. Addressing,
3. Supervision and
4. Mediation

Organizational frameworks are designed to implement these abstractions using distributed object management in human networks.

Signaling allows prioritization of the network objectives and allocates resources in the form of distributed nodes to accomplish the objectives and provides management control to mitigate risk.  Elaborate workflows are implemented using the signaling mechanism to specialize and distribute tasks to various nodes.   The nodes are used to collect information, analyze it and control themselves as a group to accomplish the required goals.  A key factor of the success of this model is the parallelism of service delivery and the service management networks using a signaling OS to manage the FCAPS element of the individuals and the group connection by leveraging the individual FCAPS management capabilities.

Thus organizational hierarchies, project management, process implementation through workflows are all accomplished through the network object model with FCAPS abstractions and signaling. It is important to note that the signaling abstractions, while commonly used, have not been discussed widely in the distributed computing domain.  First clear articulation is found in the description of SS7 signaling in telecommunications domain and a reference to it by Gartner group [6].

The comparison with human network can help us also to solve some of the problems related to authorization, authentication and trusting in distributed CPU networks.  When human baby is born, the baby is endowed with rudimentary self-FCAPS management capabilities and signaling awareness to communicate with other nodes of similar type that it interacts with (namely the mother, father, siblings etc.)  It establishes connection only after a third party authentication (Mom assures that aunt is OK to trust through signaling)

The lesson that emerges from the analogy of the human network is that each node is a computing entity with parallel computational threads that allow self-FCAPS management.  A group of nodes are able to form communities with signaling capabilities to support network composition and execute a workflow as a group.  The following video presented in SMTPS2011 Workshop at International Parallel Distributed Processing Symposium illustrates the new computing model.

Video shows a signaling OS network that allows FCAPS management of the computing network to organize the workflow execution.  A network-centric computational model requires a signaling OS that supports FCAPS management in parallel along with computing OS implementing the computational workflow.  A network of CPUs then is dedicated to orchestrate specialized tasks constituting the workflow.  The multithread architecture allows real-time implementation of both signaling OS and computing networks in parallel.

Grids, Clouds and Network-centric Computing – A Search for Unified Theory of Computing

The search continues for new programming abstractions that strive to find efficient ways to utilize computing resources (distributed or otherwise) to implement service workflows which accomplish an end goal in a reliable and secure manner with the highest performance at a lowest possible cost.  Will the network-centric computing inspired by the human networking model provide the next big break?

The Grid was originally designed as a large network of computer systems able to offer an environment where computing and storage resources are shared among academic and research institutions. To date, the development of standards, such as the Open Grid Service Architecture (OGSA), along with the introduction of Web Services technologies and new paradigms, such as the Semantic Grid, is leading the Grid toward new applications, embracing not only computational-intensive e-science applications, characterizing the scenario where grid is born and developed, but also for computing scenarios like service and information providing, multimedia environments, ubiquitous computing, etc.  On the other hand, Cloud Computing takes advantage of the new virtualization technologies and large network bandwidth to provide virtual computing resources transcending physical computing, network and storage infrastructure and geographical boundaries.

Figure 2 shows a virtual Grid in a cloud.

Virtual Grid In a Cloud

The Cloud Computing paradigm decouples the hardware infrastructure and the virtual computing infrastructure, thus offering a grid of virtual computing, network and storage resources to implement a virtual grid services network, overcoming the limitation of present grids due to their rigid schema.

Realizing the benefits of this convergence using emerging new technologies for next generation Grid and Cloud Computing is the goal for the new workshop that was formed by combing the ETNGRID (http://etngrid.diit.unict.it/) and Collaboration and Cloud Computing (www.workshop.kawaobjects.com) Workshops conducted under the aegis of WETICE2010 (http://WETICE.org/) held in TEI, Larissa, Greece.

The participants of the ETNGrid and Collaboration and Cloud Computing workshops agreed that the objectives of the new workshop named “Convergence of Distributed Computing Clouds and Grids” are two-fold:

1. Develop a simplification of the architecture to facilitate the decoupling of hardware infrastructure management from virtual services management so that the service workflow implementations are truly distributed and managed based solely on their latency tolerance and not on geographical or physical infrastructure boundaries [7, 8].  This will allow eliminating layers of management infrastructure which burdens the data centers today that has evolved from a server-centric architecture with limited bandwidth networks.
2. While both Grids and Clouds offer pooling and sharing of distributed resources, management of resources becomes a critical factor to resolve contention based on business priorities of the consumers of the shared resources.  In addition, in order to meet the wildly fluctuating demand for resources in a mass market, dynamic configuration, Fault, Accounting, Performance and security management must accompany the resource sharing infrastructure.  Therefore, Investigate ways to implement telecom grade “trust” in the cloud so that services can share resources distributed over different clouds based on their latency tolerance and business priorities.

Today’s server-centric operating systems are more suited for operating on the resources in a virtual computer in a virtual grid rather than managing distributed resources that contribute to a virtual computer in the cloud.  A distributed OS that assures Fault, Configuration, Accounting (of utilization), Performance and Security of a composition of distributed resources that constitute a virtual server is required to dynamically provision service levels such as the CPU, memory, bandwidth, storage capacity, IO and throughput on demand.  We call this OS, a Signaling Operating System (SOS)

Borrowing concepts from human networks, telecommunications networks and the Internet, a distributed computing model proposed by Rao Mikkilineni was discussed in the context of distributed resource management.

The model incorporates FCAPS management using a signaling network overlay and allows the dynamic control of the computing element with respect to its configuration, Fault management, Performance management, security management and accounting management.  Such parallel implementation of signaling and computing networks is feasible today with multi-core, multi-CPU hardware assisted virtualization.

Figure 3 shows the computing model constituting of networked computing elements called Distributed Intelligent Managed Elements (DIMEs) taking advantage of the signaling network overlay on the computing network.  Each DIME contains one or more CPUs with read/write capabilities to memory, network and a storage device.

Services and Their Management

Figure 4 shows a possible implementation example of a distributed cloud with services and their management.

FCAPS Managed and Signaling Enabled Distributed Computing with Distributed Managed Intelligent Element (DIME) Networks

The dynamic dial-up and dial-down of computing, network and storage resources with reliable FCAPS management allows a simpler architecture to implement the Grid services.  The overlay signaling control network provides a new degree of agility in FCAPS management.  The Signaling OS also provides true distributed computing that is highly scalable inside the server and also among distributed servers, and other laptop or mobile devices that support hardware assisted virtualization.

Figure 5 shows possible computing cloud architecture with signaling network overlay in parallel that allows FCAPS managed virtual server Grid network with pooled virtual resources independent of physical infrastructure or geographical boundaries.

The participants agreed to study these and other ideas to discuss their impact on Grid and cloud convergence in the next WETICE 2011 Conference in Paris (June 27^th – 29^th, 2011)

Why Focus on Dynamic Management of Resources (Physical and Virtual) is Important:

As both businesses and consumers demand communications, collaboration and commerce at the speed of light, the data centers are becoming the nerve centers of global information network.  Mobility and user interactivity using a variety of information access devices and user-centric applications have created a need for information access on-demand, anywhere and at any time.  Figure 6 shows current data center estimates of Total Cost of Ownership over a 5 year period with infrastructure that is virtualized compared to the infrastructure that is not virtualized.

Few key points to note are:

1. The TCO over a five year period is 1.87 better with virtualization with 150 servers and is 2.04 times better with 1500 servers.
2. First year investments are $1.24M and $5.36M with 150 and 1500 server data centers respectively while the first year savings are $0.57M and $2.59M respectively.
3. The infrastructure management costs over 5 years are about 40% to 45% of the total cost with or without virtualization.
4. The software and services costs are approximately 20% of the total cost.  In a data center with 1500 servers, the total cost of infrastructure management and virtualization software and services costs add up to 70% of the TCO.

The main conclusion one can draw is that there is still room for improvement to reduce the infrastructure management and virtualization software and services costs as a percentage of TCO.

Even though the TCO reduces substantially by about a factor of about 2, it is interesting to note that Infrastructure management cost remains around 40% of the total cost.  Virtualization software and services add about 20% of the total cost.

Real progress cannot be made till the management cost is substantially reduced with real-time dynamic and automated computational workflow management.  Is it time to reexamine the current server centric computing model that has evolved over decades when bandwidth abundance and multi-core architectures were not the norm?  Parallelism in real-world is the norm.  Will the Nature Inspired Computing (NIC) models be the new norm in future computing?  Whatever the case may be, current economic trends cannot support a 40% management cost and 20% software and services cost in the future data centers.  Out of box approaches are essential to create a paradigm shift to match the 10X improvement in hardware innovation with equal gains in software innovation.  Will the network-centric computing model allow hierarchical scaling and parallelism with local autonomy and higher level coordination and supervision based on policies?  Is signaling crucial for providing the agility to the workflow implementations in a network of heterogeneous nodes?

The next conference of WETICE2011 in Paris will include a session on the new paradigms that will enable next generation fully virtualized data centers to exploit multi-core and multi-CPU systems, signaling OS alternatives, parallel programming technologies and highly available, reliable, secure and high-performance workflow management architectures.   A call for papers will be forthcoming soon.

References:

[1] Edited by Keckler, Stephen W., Olukotun, Kunle, Hofstee, H. Peter, “Multicore Processors and Systems” Springer, 2009

[2] David Patterson, “The trouble with multi-core”, IEEE Spectrum, July 2010, p28

[3] http://www.thesystemsthinker.com/systemsthinkinglearn.html

[4] http://www.orgnet.com/MCO.html “Managing the Connected Organization” by Valdis E. Krebs

[5] Thomas W. Malone, “Organizing information systems: Parallels between human organizations and computer systems”, Cognition, Computing and Cooperation, Cognition, Computing and Cooperation, edited by Scott P. Robertson, Wayne Zachary, John B Black, Greenwood Publishing Group, January, 1990

[6] Gartner Dataquest Says Next Generation Signaling to Prosper with Emergence of Next Generation Networks, Business Wire,  Feb 12, 2001

[7] Vijay Sarathy, Purnendu Naraian, Rao Mikkilineni, “Next generation Cloud Computing Architecture – Enabling real-time dynamism for shared distributed physical infrastructure”, Proceedings of IEEE WETICE 2010, Second International Workshop on Collaboration & Cloud Computing, June 2010 (http://www.kawaobjects.com/resources/PID1258479.pdf )

[8] Rao Mikkilineni, Vijay Sarathy ”Cloud Computing and Lessons from the Past”, Proceedings of IEEE WETICE 2009, First International Workshop on Collaboration & Cloud Computing, June 2009 (http://www.kawaobjects.com/resources/ieee-ccc-2009.pdf)

The combination of hardware assisted virtualization and the broadband Internet have taken the Information Technology (IT) hosted managed services to a next level of evolution, where the software applications have become independent of the hardware infrastructure and can be migrated at will.  This introduces two key issues that need to be addressed to fully leverage the potential that the new servers and virtualization offer:

1. The operation and management of virtual services have to be decoupled from the operation and management of the server, network and storage physical infrastructure that hosts them
2. Resource provisioning must be application sensitive, dynamic to meet the transient nature of services that can migrate and must accommodate latency constraints of services that utilize them.

The second international IEEE Workshop On collaboration & Cloud Computing under the aegis of  19th IEEE International Workshops on Enabling Technologies: Infrastructures for Collaborative Enterprises (WETICE) 2010, is devoted to address the operation and management issues in the next generation cloud computing.  Seven papers discuss various aspects of bringing telecom grade “trust” and global interoperability to distributed collaborating computing clouds.

Theme: Computing Clouds with Telecom Grade “Trust” and Global Interoperability

“During the past 15 years, a continuing trend toward IT industrialization has grown in popularity as IT services delivered via hardware, software and people are becoming repeatable and usable by a wide range of customers and service providers,” according to Daryl Plummer, managing vice president and Gartner Fellow. “This is due, in part to the commoditization and standardization of technologies, in part to virtualization and the rise of service-oriented software architectures, and most importantly, to the dramatic growth in popularity of the Internet.” [1]. Gartner, Inc., the information technology research and advisory company predicts that by 2012, 20% of businesses will own no IT assets.  For these businesses, the computing resources will become a fifth utility like water, electricity, gas, and telephony offered by service providers assuring reliability, availability, performance and security.  The economies of scale will allow the service providers to offer ubiquitously, cost-effective services on demand and pay-per-use.  The computing utility is more close to the telephone utility which has evolved from being just voice connectivity to voice, data and video connectivity with the advent of the Internet. 

The Internet provides the connectivity between IP (Internet Protocol) devices to transfer voice, data or video bits at a certain level of quality assured by the service provider.  In the case of computing utility, the service provider must provide the user with access to computing, network and storage resources (CPU, memory, bandwidth, storage capacity, throughput and IO per second) with specified, reliability, availability, performance (including latency) and security constraints.  Recent cloud computing offerings from various service providers attempt to provide the computing utility service over the Internet.  What distinguishes the cloud computing services from hosted managed computing infrastructure access over the Internet is the cloud’s ability to dynamically dial-up and dial-down the resources on demand to meet the changing requirements of the services that utilize them.  This is made possible by the virtualization technologies that allow abstraction of computing resources such that a single physical machine is able to function as a set of multiple logical Virtual Machines (VMs) [2, 3]. 

A new class of servers offers three compelling reasons to refresh current data centers to take advantage of virtualization and implement public, private or hybrid clouds:

1. Energy cost reduction,
2. High performance using multi-core, multi-CPU processors with high bandwidth throughput and
3. Hardware assisted virtualization enabling service agility, high availability and disaster recovery

While the energy and space savings and high performance provide impressive return on investment by themselves, what is perhaps being overlooked is the disruption that hardware assisted virtualization could potentially usher in by truly enabling the hardware infrastructure operation and management to be completely decoupled from business process implementation, operation and management.  Virtualization allows the composition of logical servers with a choice of service levels for computing, network and storage resources independent of physical infrastructure composition or physical location.  Only a fraction of the potential has been so far realized – thanks largely due to the various Operating Systems and software-based hypervisors which provide multi-tenancy capabilities resulting in significant power and space savings through the consolidation of multiple applications on the same hardware.  In addition, the ability to migrate virtual applications from one physical infrastructure to another has radically transformed the way high availability and disaster recovery are implemented.

However, as one of the papers in this workshop points out the real potential of virtualization is not fully exploited for the following reasons:

1. Current generation Hypervisors (the operating systems that provide computing, network and storage resource management for applications that use them) have evolved from current server-centric operating systems and do not allow a clean separation of hardware Fault, Configuration, Accounting, Performance and Security (FCAPS) management from services FCAPS management to fully exploit the hardware and software decoupling.
2. Each OS vendor, in order to claim dominance with their existing products, offer other OSs as “guest OSs” without providing same level integration services across multiple OSs.  This makes the real life implementations, very management intensive, cumbersome and error-prone to really operate in a heterogeneous environment. 
3. Current method of using Virtual Machine Images that include application, OS and storage images does not provide on-the-fly composition and decomposition of application services.
4. More importantly, current server implementations of Hypervisors do not exploit the potential of sharing distributed resources across physical and geographic boundaries and provide latency based composition of logical servers to fully utilize the power of distributed resources.

True power of network-centric computing (in contrast to server-centric computing) will be unleashed only when the operating system provides latency sensitive FCAPS management of distributed server, network and storage resources.

There are two schools of thought that are emerging to shape the future course of virtualized datacenters:

1. The Incrementalists:  Most vendors who have existing products would like the incremental approach to add additional layers to support virtualization.  They usually provide minimal support to other vendor products and offer “put the other guy’s system behind my system for management” approach.  Windows support of Linux and Linux support of Windows fall in this category.  While they pretend to support other approaches, they will tend to enhance their own approach to show differentiation.  Their main argument is that customers have embedded-base and do not like fork-lifts.  The Incrementalists offer incremental layers of management systems, bridge systems and a plethora of services to assist their customers implement their solutions.  They show how their point solutions incrementally improve the ROI
2. The Disruptionists:  Most startups who do not have vested interest in existing solutions attempt to show how disruption can bring orders of magnitude cost savings by eliminating many of the existing point solutions.  They offer fully decoupled services architecture in which hardware becomes commoditized and all services are virtual applications.  They claim that the new multi-core multi-CPU servers, high bandwidth SAS storage, and gigabit Ethernet offer new simplification that can eliminate a host of complex server, storage and network management systems by unifying the logical resource management.  The dynamic provisioning that virtualization offers can be used to eliminate the current system administration paradigm that has evolved over last few decades from a server-centric IT.  By leveraging the decoupling of physical infrastructure from logical servers and implementing latency based dynamic provisioning of resources, new application architectures are possible that dramatically simplify the implementation of business workflows with greater agility.  They envision that the orders of magnitude productivity improvements will naturally lead to a refresh of current data centers and provide an impetus for changing application architectures.  They point to current efforts by SAP and Oracle to reengineer their applications to meet the new drive towards offering these applications as Software as a Service (SaaS).  They claim that the disruption of the vendors is a good thing without disrupting the customer businesses.  They also assert that a seamless migration is possible with the use of dynamic provisioning that virtualization offers.

Be that as it may, virtualization indeed offers a dramatic opportunity for improving the next generation data center productivity.  How we get there will be decided by a few innovators in the next few years.  One clear lesson we can take away from POTS (Plain Old Telephone Service), the Internet (enabling Pretty Amazing New Services, PANS) and SAN implementations of the past is that when resources are distributed and shared, they need to be managed to compose them and provide services to their consumers with appropriate end-to-end service level assurance.   Service consumers and resources require mediation to resolve contention for shared resources and resolve conflicts based on global business priorities.  Services that enable communication, collaboration and commerce at the speed of light, also demand dynamic resources and services management at the speed of light.  Current server-centric architectures with their system administrator oriented design, albeit automation of administration tasks, fall short in addressing the new virtual server based services management.

The CCC workshop presents some of these issues and new approaches from both the Incrementalists and the Disruptioninsts.  The discussions in the workshop will hopefully lead to some innovative hybrid solutions.  There are seven refereed papers addressing these approaches.

According to Gartner’s report on Hype Cycle for Cloud Computing, 2009, “Cloud computing is the latest super-hyped concept in IT. However, it is simplistic to look only at the hype around the high-level term. As aspects of the cloud move into mainstream adoption, there will be misunderstandings, overestimations and underestimations that will cause users to have doubts and be disillusioned. There will be misuse and miscommunication among users and vendors, making it a subject that is ripe for Gartner’s Hype Cycle.  As usual, in cases like this, there is indeed overhyping, but also benefits. Understanding those benefits requires tearing apart the hype surrounding cloud computing (just at the Peak of Inflated Expectations, and headed for the Trough of Disillusionment), and looking at the many more-granular topics, which are all part of the cloud phenomenon. This follows the pattern observed with other similarly broad labels as “the Internet” and “the Web.”  [4]

The 10X improvement in performance offered by the new multi-core, multi-CPU servers will drive a major refresh of the datacenters to benefit from consolidation, performance improvement, instant-HA/DR capabilities, and service agility.  However, the hardware assisted virtualization offers an architectural simplicity that will decouple the operation and management of hardware infrastructure from the operation and management of the services implemented in the virtual infrastructure.

While current state of the art of cloud computing has not yet assured confidence with telecom grade trust, if approached correctly, I believe that computing clouds present the last frontier in information technology revolution that will reduce complexity, provide global interoperability and telecom grade trust to networked computing resources.  Based on POTS, PANS and SAN experience, it is easy to see that reducing waste and providing telecom grade trust for accessing computing resources globally will have profound economic consequences.

The key to a paradigm shift is to recognize that the current administration and management paradigm that originated with server architecture is static and assumes that the resources (CPU, memory, bandwidth, storage capacity, throughput and IOPs) are allocated to an application at install time.  Changes to workloads and business priorities are assumed to occur at longer time scales, and administration can be performed off line.  This assumed that there were maintenance and administration times that are scheduled and the services the application provides can be interrupted during this period.

In a virtual data center dynamic provisioning must be application sensitive, dynamic to meet the transient nature of services that can migrate and must accommodate latency constraints of services that utilize them.

The Second international workshop on collaboration and cloud computing addresses some of the issues associated with current cloud architectures and also explores some new ideas to radically improve the impact of public, private and hybrid clouds in implementing robust business processes with telecom grade “trust” and global interoperability.

Venue:

WETICE 2010 will be hosted from June 28th to June 30th, 2010, at the:

     DDE Building
     T.E.I. of Larissa
     Zip 411 10,
     Larissa, Greece

WETICE is an annual international forum for state-of-the-art research in enabling technologies for collaboration, consisting of a number of cognate workshops.  WETICE-2010 is co-sponsored by the IEEE Computer Society Technical Committee on Data Engineering.  So far 18 WETICE conferences have taken place under the co-sponsorship of IEEE CS.  What sets WETICE apart from larger conferences is that the workshops are kept small enough to promote fruitful discussions on the latest technology developments, directions, problems, and requirements.  Each workshop includes paper presentations, workgroup discussions, keynote sessions and a final joint session to summarize each groups’ findings.  The Proceedings will be published by IEEE CS Press and distributed at the conference.  WETICE 2010 includes also plenary invited presentations by experts from academia and industry and a Young researchers workshop on Services particularly tailored at PhD students.

For details on how to register, please visit http://WETICE.org 

References:

[1]     http://www.gartner.com/it/page.jsp?id=707508

[2]     P. Barham, B. Dragovic, K. Fraser, S. Hand, T. Harris, A. Ho, R. Neugebauer, I. Pratt, A. Warfield, Xen and the art of virtualization, in: Proc. 19th ACM Symposium on Operating Systems Principles, SOSP 2003, Bolton Landing, USA, Oct. 2003.

[3]     Rajkumar Buyyaa, Chee Shin Yeoa, , Srikumar Venugopala, James Broberga,  and Ivona Brandicc, “Cloud computing and emerging IT platforms: Vision, hype, and reality for delivering computing as the 5th utility”, Future Generation Computer Systems, Volume 25, Issue 6, June 2009, Pages 599-616

[4]     Gartner RAS Core Research Note G00168780

The combination of hardware assisted virtualization and the broadband Internet have taken the Information Technology (IT) hosted managed services to a next level of evolution, where the software applications have become independent of the hardware infrastructure and can be migrated at will.  This introduces two key issues that need to be addressed to fully leverage the potential that the new servers and virtualization offer:

1. The operation and management of virtual services have to be decoupled from the operation and management of the server, network and storage physical infrastructure that hosts them
2. Resource provisioning must be application sensitive, dynamic to meet the transient nature of services that can migrate and must accommodate latency constraints of services that utilize them.

The second international IEEE Workshop On collaboration & Cloud Computing under the aegis of  19th IEEE

International Workshops on Enabling Technologies: Infrastructures for Collaborative Enterprises (WETICE) 2010, is devoted to address the operation and management issues in the next generation cloud computing.  Seven papers discuss various aspects of bringing telecom grade “trust” and global interoperability to distributed collaborating computing clouds:

1. Next Generation Cloud Computing Architecture: Enabling Real-time Dynamism for Shared Distributed Physical Infrastructure
2. A Framework of Scientific Workflow Management Systems For Multi-cloud Orchestration Environment
3. Application Development: Fly to the Clouds or Stay In-house?
4. Using Virtualization to Prepare Your Data Center For “Real-time Assurance of Business Continuity”
5. CloudGauge: A Dynamic Cloud and Virtualization Benchmarking Suite
6. Enterprise Usability of Cloud Computing Environmenr: Issues and Challenges
7. Creating Next Generation Cloud Computing Operations Support Services By Social OSS Contributions With Telecom NGN Experience

WETICE 2010 will be hosted from June 28th to June 30th, 2010, at the:

     DDE Building
     T.E.I. of Larissa
     Zip 411 10,
     Larissa, Greece

WETICE is an annual international forum for state-of-the-art research in enabling technologies for collaboration, consisting of a number of cognate workshops.  WETICE-2010 is co-sponsored by the IEEE Computer Society Technical Committee on Data Engineering.  So far 18 WETICE conferences have taken place under the co-sponsorship of IEEE CS.  What sets WETICE apart from larger conferences is that the workshops are kept small enough to promote fruitful discussions on the latest technology developments, directions, problems, and requirements.  Each workshop includes paper presentations, workgroup discussions, keynote sessions and a final joint session to summarize each groups’ findings.  The Proceedings will be published by IEEE CS Press and distributed at the conference.  WETICE 2010 includes also plenary invited presentations by experts from academia and industry and a Young researchers workshop on Services particularly tailored at PhD students.

For details on how to register, please visit http://WETICE.org

“There is nothing more difficult to take in hand, more perilous to conduct, or more uncertain in its success, than to take the lead in the introduction of a new order of things.”                      –Niccolo Machiavelli, “The Prince”  1469-1527

In spite of Machiavelli’s warnings, human endeavor strives to introduce a new order of things time and again.  This is more so in technology than in any other human undertaking.  Current industry effort to transform the Information Technology delivery using virtualization is a case in point.  This quote from a startup company Vyatta spells out the future that the new hardware assisted virtualization in the multi-core multi-CPU servers are offering to fully transform the nature of the next generation data center.

“Vyatta is bringing innovation and affordability to the networking industry by delivering advanced routing and security in a software-based network OS that scales from the branch office to the service provider edge. Vyatta has decoupled networking software from proprietary hardware allowing users to leverage the price and performance advantages of standard x86-based hardware and components as well as Citrix XenServer and VMWare virtual or cloud environments.”

We have seen the sunset networking technologies before.  The current generation technologists probably do not know what an analog or step-by-step switch is or that ATM once stood for next generation broadband technology.  While IBM still continues to support SNA, many of the people who can spell out what SNA stands for are either retired or dead.  As the following quote from Wikipedia points out, a vestige of legacy still continues.  “While IBM is still providing support for SNA, one of the primary pieces of hardware, the 3745/3746 communications controller has been withdrawn from marketing by the IBM Corporation. However, there are an estimated 20,000 of these controllers installed and IBM continues to provide hardware maintenance service and micro code features to support users. A robust market of smaller companies continues to provide the 3745/3746, features, parts and service.”  The rest of the world moves on.

As Vyatta correctly points out the decoupling of virtual services from physical hardware infrastructure offers a new simplification of the next generation virtual data center by eliminating a host of proprietary appliances, protocols and siloed management systems.  Elimination of heterogeneous cable tangling and the interoperability issues with multiple protocols, that resulted from decades of overlay network evolution, is made possible by the “virtual network” inside the server smoothly integrating with the broadband transport network outside the server.  It offers a paradigm shift with multi-fold reduction in Total Cost of Ownership (TCO). 

However what is not appreciated is that the new infrastructure and advances in virtualization technologies offer a new distributed workflow implementation with telecom grade trust where self configuring, self monitoring and self-optimizing services collaborate to offer latency sensitive cpu-to-spindle connections on demand.  Instead, many of the efforts today are going toward consolidating existing applications, which are tied to physical infrastructures that are proprietary and burdened with layers of administrative systems,thus contaminating the simplicity of the new architecture.  Is this “old wine in a new bottle” approach (perhaps square pegs in round holes is a better metaphor) the right approach to realize the orders of magnitude improvement that the new generation of servers offer? A clear example is running a current Fibre Channel dependant application as a virtual application.  Imagine the complexity that will be ported inside the server when two Fibre channel based applications with multipathing are implemented in a virtual server.  Performance optimization in such an environment will be a nightmare.  Even Fibre Channel over Ethernet attempts seem artificial and run counter to choosing the best architecture that exploits virtualization with management simplicity and agility using dynamic provisioning.

The raw chip performance that is tenfold along with hardware assisted virtualization of CPU and memory in the new server architecture offer an unparalleled opportunity to improve the way business workflows are implemented.  They offer an opportunity to eliminate decades of layered appliance complexity with multiple protocols with their own interoperability and management overhead.  For the first time, it is possible to replace myriad point solutions that have been strung together in the guise of “open systems” with next generation integrated solutions for the cost of maintenance of today’s systems alone.   It is estimated that 80% of current IT budget goes to maintaining existing systems than creating new services to serve the business needs. 

There are two key issues that need to be addressed to fully leverage the potential that the new servers and virtualization offer to avoid putting old wine in the new bottle:

1. The operation and management of virtual services have to be decoupled from the operation and management of the server, network and storage physical infrastructure that hosts them
2. Resource provisioning must be dynamic and must accommodate latency constraints of services that utilize them.

Current server-centric operating systems that have evolved over few decades, where dedicated resources were assumed and the resources are statically provisioned to applications at install time, no longer fit the new model where services can be dynamically provisioned on-demand and shut-down when not in use.  New operating systems that provision distributed resources on-demand to match the latency tolerance of services that use them are required that go beyond today’s Hypervisors.  Distributed resource Fault, Configuration, Accounting, Performance and Security management has to be integrated with services and service FCAPS management with appropriate mediation to enable distributed self-managed business workflows that are latency tolerant.  Just automating existing administration processes using scripts to bridge current siloed management systems will not suffice. Leveraging hardware assisted virtualization  using next generation distributed operating system offers an opportunity to eliminate them.  Will the architectural simplicity of distributed computing clouds with hardware assisted virtualization trump the status quo of management complexity or the square pegs in round holes prevail?

The 2nd International IEEE Workshop is intended to continue the discussion of some of the innovative approaches started in the last workshop by bringing both the academic researchers and industry R&D leaders together.  Papers are invited to discuss innovative approaches that will enable next generation virtual data centers that will enable distributed collaborating computing clouds.

The last date for submission of papers is extended to:

March 14^th, 2010.

Please submit your papers to workshop@kawaobjects.com

For details on how to contribute papers and how to participate in the workshop, please visit

www.wetice.org

www.workshop.kawaobjects.com