The Missing Link

It’s easy to see why the boffins are so excited. Enabling different computing systems to virtually share applications and data, grid computing harnesses the power of hundreds or even thousands of servers at geographically separated installations – such as branch offices – ultimately allowing applications to run at far greater speeds.

Much of the legwork of this innovation is done by software packages that analyze the resources available across a network, and applications that then divide this up and parcel it out to the servers comprising the Grid. Put in simple IT terms, it utilizes the wasted storage capacity that languishes in most organizations’ servers, gathering distributed resources of CPU power, storage capacities, network bandwidth, printers, people and instruments, and providing the collective to those who need it. As such, users simply need to plug into a hive of unified power and draw upon resources that were previously unobtainable.

It’s impressive stuff. However, the essence of grid computing has been in existence for some time, so are we any nearer to achieving the virtual dream? BM asks the experts:

Ken King, Vice President of Grid Computing at IBM
Mark Linesch, Chairman of the Global Grid Forum
Yaron Benvenisti, CEO of GigaSpaces
Peter Lee, Co-Founder and CEO of DataSynapse
Enrique Castro-Leon, Principal Enterprise Architect at Intel Solution Services, Intel Corporation

BM. Is it true that grid-based enterprise IT infrastructure is the next logical evolution in IT?
YB. It’s very accurate, maybe even an understatement. Grid architecture is a tremendous leap in the evolution of IT. When the vision materializes, it will mean business applications can scale on demand when and where they need to, a chieve optimal performance, and become inherently fail-safe – when units break down, others can pick up the load without interruption. These have been the perennial challenges of application development and deployment for as far back as we can remember. Grid will render them obsolete.

KK. The real evolution in the marketplace will be towards a more distributed computing model, giving customers the ability to run their applications and business processes more effectively across heterogeneous systems. Grid is just a set of differentiating technologies that fits well into that distributed computing model. I see a significant move towards services orientated architectures (SOA) and applications being written as services based applications that are more modular, componentized and provide more flexibility as to how the application is run within a customer environment. The infrastructure underlying those applications and business processes is going to be more dynamic as well. In the short-term, Grid is continuing to develop out its capabilities and will get a lot of hype over its ability to create that dynamic infrastructure and help accelerate application performance. But over time, I think those core capabilities will evolve and start commoditizing into base system software and core middleware. I do see Grid as being part of the next logical evolution of IT but not by itself. It is a core catalyst for helping get us there.

ECL. Grid computing represents the next logical step for server virtualization. The most often cited advantage is increased utilization levels, and hence reduced capital expenditure. This economy could be overrated because the cost of a server may be a minimal portion of the cost to build and maintain a data center. An underestimated factor is that server virtualization enhances operational excellence and increases business agility. A virtual server can be allocated in matter of minutes, whereas a new physical server may have to go through a procurement cycle that could take as long as a year. This trend will accelerate in a grid environment: an application might need 5,000 nodes to run but these nodes might be needed only a few days per year. Spending the several million dollars needed to purchase, install and configure this many nodes is out of question, even for large organizations. Instead, these computations could be spread out to many nodes rounded up around the globe and started in a matter of minutes. The bottom line is that grids will bring scalability and dynamic behaviors hard to imagine today, increasing business agility and perhaps realizing the nirvana of utility computing.

PL. More than just being the next logical step, we believe grid-based IT infrastructures are the next inevitable step in the evolution of IT systems. Grid computing represents a next-generation paradigm shift in IT infrastructure, delivering substantially higher service levels at radically lower costs, and enabling organizations to immediately respond to business change. By eliminating inflexible, under-utilized application silos; grid computing supports enterprise movement toward SOA, which delivers loosely-coupled and agile IT services. Grid-based computing also achieves scalability on demand as applications can automatically be shared via a virtualized computing infrastructure.

BM. How have companies adapted the Grid to their business needs and addressed initial implementation challenges?
KK. GC was primarily used to share information across multiple universities, more effectively leveraging their computing resources and treating multiple physical devices as if they are one logical supercomputer. Business started off using grid in this way also by, for example, running complex simulation algorithms in the automotive industry and for core research in drug discovery. More and more we are starting to see it used in other ways, such as leveraging multiple applications and being able to run them across computing infrastructure, using scheduling software to take several different applications within a business application and put them on the grid. This lets the grid decide how best to utilize the assets and the IT resources to run the applications, based on business policies. We are also seeing more data grid implementations. A recent customer used our information integration software across multiple regional banks to be able to have one logical view of their loan data and cut out a significant amount of paperwork. It has cut down the loan processing time by 40 percent.

YB. We are seeing Grid being approached in two ways: the first is a more ‘traditional’, top-down mandated Grid effort by a central technology group, where the primary motivation is optimizing CPU usage across the enterprise. The more interesting, newer one is being driven by specific application groups that are tackling scalability and performance challenges in their systems. They are turning to Grid because it provides them with the most simple, elegant and cost-effective solution. The next step in the adoption of Grids will happen when senior IT executives realize that this top-down and bottom-up approaches are one-and-the-same, and need to be consolidated in a more systematic way. We are seeing this new pattern of adoption happening in most of the major financial exchanges, investment banks, cellular carriers and defense organizations, where the pressure to constantly achieve better scalability, faster processing and fail-safe reliability is the main competitive advantage IT can deliver.

PL. Today, we have over 40 global enterprises using our grid computing technology in production, when we had just one client four years ago. Typically, companies focus on grid-enabling their business-critical applications which require scalability, such as decision support applications that are compute-, data- or transaction-intensive. This is where they most radically reap the benefits.

ECL. We are just at the beginning in terms of universal adoption. Critical mass will not happen without two essential ingredients: standards that allow accessing resources in a uniform manner across the globe, and a number of foundational technologies to manage these resources in a secure environment that preserves the integrity and privacy of data. With standards in place we should see acceleration in adoption with lessened requirements in terms of technical sophistication.

BM. What trends do you see today in the way that GC is used?
ML. The growing trend in research and business is to focus horizontally within and across organizations for collaborative research and design; process and data integration. This trend has been accelerating through the availability of low cost, standardized infrastructure and the widespread adoption of the internet and internet-related standards for information, collaboration and commerce. Research at corporate labs, government and major universities is increasingly a ‘team sport’, where collaborative investigation and problem solving are the norm and the sharing of compute and data resources is required. Business has embraced the concept of the ‘extended enterprise’, integrating partners, suppliers and customers into dynamic value chains. The ‘extended enterprise’ places increased demands on business processes and the supporting IT infrastructure to work in a synchronized manner across a heterogeneous, networked environment. Both research and business are highly motivated to continue to support efforts to break down existing technology, process, and people silos that inhibit the flow of information, innovation and commerce in our increasingly interconnected economy. As organizations use grids to ‘virtualize’ their internal IT infrastructure and their external connectivity, they find that they can adapt more easily to these trends.

PL. Our early grid computing initiatives focused almost exclusively on compute-intensive calculations. However, as we’ve expanded deployments throughout our client organizations we have seen a much wider range of data- and transaction-intensive applications exploiting the benefits of this technology. For example, we have several clients who have deployed our technology over thousands of CPUs and have multiple applications sharing this infrastructure. Such success drives depth and breadth of grid adoption. Also, we now find ourselves working around the clock with many ISVs who – as a result of our client overlap – are coming to us to grid-enable their applications in order to create a win-win-win joint value proposition with mutual clients. The key drivers of grid adoption, however, have remained the same over time – substantially higher service levels at radically lower costs.

YB. Much of the new application development in the high-end business-critical systems (where low-latency of information, high-volume of data, and fail-safe requirements are non-negotiable) is turning to Grid as the underlying architecture they want to implement in order to achieve these goals. This is compared to the last couple of years when Grid was primarily adopted by the off-line, compute-intensive type of systems. The benefits were focused more on Grid’s value in parallelization of processing, with the resulting shortening of processing timeframes, and optimizing the usage of the underlying computing power. We are seeing more and more development using Grid concepts, where the value being sought is linear and dynamic scalability, with guaranteed reliability through a much more simplified and elegant approach.

ECL. Migrating applications to run in a grid environment is not trivial. It is at least an order of magnitude harder than migrating an application from a uni-processor environment to a multi-processor and/or multi-core environment. Considering how difficult it is to have a vendor recompile an application, it is fair to say that not every application will be modified to run in a grid. For some applications, such as productivity applications, doing so may not make sense at all. It makes sense to start with the most computing intensive applications, such as Business Intelligence or Data Mining – the ones that can take overnight if run on a single CPU. Grids originated from initial scientific and academic research for running high performance computing applications. We will see a gradual diffusion into other industry verticals as the technical barriers are tamed. It is at this point where the economic benefits of grids will be fully realized. I can see a slightly faster adoption in an Open Source environment, if only because the availability of source code invites experimentation.

KK. Over time, we’ve started to see efforts towards enterprise optimization. Lets say you have an application which, if it goes down, you have automatic revenue loss. With grid it can sense if the hardware center goes down in the data center running that critical application and can reallocate that application to other resources and get that critical application running again within its SLA. That’s the resilience you can get out of grid computing that really drives the incremental value of your business. We see enterprises testing it within individual lines of business and seeing the value of that, and then figuring out how to extend that across the enterprise. That’s challenged today with the politics about giving resources up to the Grid to be used by other applications that didn’t pay for them – that’s a challenge those organizations have to get past before they can get true enterprise optimization from grid technologies

BM.What should organizations look for in their grid solution or vendor?
ECL. Hardware is cheap, everything else is expensive. Beware of purely technical approaches. Don’t look at server utilization as the single, overriding figure of merit. It is essential to take on a holistic approach, including operational factors and the impact on the cost of labor. The per CPU licensing schemes from many software vendors is too static and cumbersome and will have to change in a grid environment. It does not make sense to purchase 5,000 nodes to use for a few hours per year and even less sense to purchase 5,000 application licenses to support a few critical large runs, when most of the runs are done on 10 CPUs or less. Ironically, if this situation does not change it might open opportunities for grid service providers who can average the cost of purchasing a large number of licenses over clients in multiple companies.

PL. Seek out a grid solution that is enterprise-ready with the capability to address the widest universe of in-house and third party applications, maintain high QoS and SLA levels, and operate in heterogeneous environments. Ask to see proven examples of production success, including extensive benchmarking of the qualitative and quantitative service benefits and cost reductions. Finally, great care should be taken in examining the repeatability of an enterprise solutions methodology in order to ensure the solution ultimately fits the client’s reference architecture and can be deployed with appropriate, expert vendor support.

YB. Address the promise and challenge of Grid in a more strategic way – more specifically as an enterprise-wide, systematic approach to building and architecting all of their core business systems moving forward.

There is a convergence of technologies on the go, where the ‘traditional’ processing grid technologies (i.e. Compute Grids), data distribution technologies (i.e. Data Grids) and SOA (i.e. Service Grids) are coming together as a common platform for building and deploying Grid-enabled applications. This will not be provided at any time soon by a single vendor but there are already some key players in each space that are taking steps to integrate their products and ensure their interoperability. It is crucial that organizations understand Grid can’t be bought, you have to build it.

Identify key players and understand that to maximize the value of these disruptive concepts the infrastructure being built should be enterprise-wide and cross application. As such, the vendors behind it need to prove their products are mature, can point to successes and, most importantly, are built on open-standards rather than proprietary ‘black-boxes’.

KK. Start small and identify an application area that is not getting good utilization of its assets and had issues with meeting SLAs. It’s important to look for a place where you can address a pain point specifically within your business and get a quick return on that, so that you can then use that pilot to justify extending it to other parts of your enterprise. Look for a vendor with expertise that has implemented grid computing previously to get quick ROI from working with that vendor. Integration is key and understanding how to integrate scheduling and provisioning technologies with information, integration and federation technologies is a core element of picking a vendor who knows how to build grid computing solutions. Choose one that has a technology or solution roadmap and is trying to solve business issues rather than just trying to be a technology provider. Lastly, you need a vendor that works closely with the grid standards community who is focused on open-based grid solutions. If they are focused on driving proprietary implementations, you will be locked into that vendor and won’t be able to build this out extensively over your enterprise.

ML. The key to realizing the benefits of grid computing is standardization. This is so the diverse resources that make up a modern computing environment can be discovered, accessed, allocated, monitored and, in general, managed as a single virtual system – even when provided by different vendors and/or operated by different organizations.

BM. Are there limitations? What realistic ROI should companies expect?
KK. It’s very hard to give a specific ROI statement but we have seen cases where customers have seen 50 percent hardware cost savings from their grid implementations. One of our clients, Mass Mutual, stated at the Grid Today VIP Summit earlier this year that they have calculated a 60 percent reduction in cost over three years with their gird implementations. Beyond the hardware saving, there is the broader ROI that you really can’t calculate in terms of savings per se, for example, the ability to solve complex business problems that would be hard without grid technology. Another example in accelerating application performance is Magnus Dyer, an auto parts supplier in Europe. They had a clash algorythm and used grid technologies to cut the processing time from 72 to four hours. By doing so they were able to take that clash algorithm that they were running at the end of the design process and run it nightly. This enabled them to change the whole way they ran their design process, increase the quality and accelerate the time to market; giving them a real competitive advantage.

ML. The point where companies run into trouble is not taking the time to fully explore what applications and processes they want to run on a grid prior to adoption, and developing the right measures to compare with their investment to determine the return. For example, Royal Bank Insurance shortened the time needed for certain actuarial calculations from 18 hours to 32 minutes. RBC Insurance reduced time spent on job scheduling by 75 percent and time spent on processing actuarial applications by 97 percent. General Motors has deployed a grid to help simulate vehicle crash analysis with the goal of cutting an estimated 100 – 200 vehicles from physical crash testing, saving approximately US$300,000- 500,000 per crash test. I participated in a panel during the Fall GridWorld event discussing the subject and participants who have deployed grids consistently remarked that grids are not huge investments, especially if you start small and grow progressively. Most of the adopters I have talked to have realized ROI over a period of weeks and months, not years.

ECL. Grid technology allows enormous flexibility in trading off between capital and operational expenses, allowing organizations to hit the best balance for them. Short term, some gains will be attained through resource substitution, i.e. a given task could be executed in a fraction of the original time. The real quantum jump in productivity will take place when business processes get re-engineered to take advantage of grids.

Long term, expect a multiplier in ROI, not because of the capabilities the grid brings but because of the new business processes and relationships it enables. For instance, if data and programs can be moved at a very low cost and secured to the point that data is inscrutable by host providers, it’s reasonable to expect that hosting data centers could migrate to low cost of labor countries. This could be a highly disruptive change.

YB. One of the immediate values that can be gained from adopting Grid as a solution for scalability and performance is the short-term ROI associated with being able to use low-end commodity hardware platforms, like blades, running Linux as the operating system. Just recently, business critical, high-performance and transaction-intensive applications could only run on the very high-end of server technology. We have seen this ROI materialize overnight for some of the major financial exchanges and investment banks in the US, and for Telecom vendors like Nortel and Hutchison 3G.

PL. Our clients have consistently achieved orders-of-magnitude improvements in service levels, defined in terms such as application performance, resilience, or scalability; while slashing the costs of development and production support. Moreover, our clients have documented annual capital cost reductions of one-third to over one-half of their previously budgeted costs, and annual operating expense reductions of around one-fifth of their budget. Sharing, automation and ease of management deliver real and immediate business benefits. Most notably, our clients consistently cite that the cost benefits are really icing on the cake. The qualitative benefits of being able to immediately respond to business unit requirements, as well as providing substantially higher service levels, are estimated to be 3-5 times more valuable to the business than the cost savings alone.

BM. What uncapped potential do you see for the Grid. Are companies realizing the full benefits?
ECL . This is the $64,000 question. With grids, we are in a stage similar to radio transmission in the early 1920s. Some thought of this technology as a “better” telegraph or a “better” telephone because it required no wires. In retrospective, this thinking was too constrained, based on the experience of the time. The industry did not take off until the Radio Corporation of America (RCA) came up with a broadcast model, using the radio to reach large audiences and as an advertising medium. It is possible that the usage models that will lead to pervasive adoption of grids are yet to be discovered. A certain level of risk has already been factored into early grid deployments. This will encourage early adopters to look at the grid as a microcosm for the larger enterprise computing and use it as a test bed for emerging concepts such as SOAs and advanced IT management technologies. The experience from these deployments, including the operational implications of dynamic resource allocation, will put the early adopters in a strong competitive position, even if grids do not become ubiquitous in their shops. This experience will be taken to enterprise computing at large, completing the full circle for grid computing.

YB. We are just scraping the surface of this revolution. Grid concepts, coupled with their tightly coupled ‘older-brother’, SOA, will change the economics of business applications. Organizations, including ISVs, that realize this approach is here to stay and adopt it across their IT will have the edge. While the challenge to adopting this is formidable, given the investment in current systems and installed base, there are already proven technologies out there which can be used to get started with this effort. For this to happen, it is imperative that CIOs and heads of IT realize that in the long run, Grid computing is their ONLY ticket for getting their information systems to scale up to what their businesses require in terms of ability to process more and more information in ever decreasing timeframes – all this with 100 percent reliability and with much lower costs.

PL. Our clients are aggressively exploiting the benefits of grid computing but the specific pace of a production rollout is subject to the particular objectives, resourcing issues and other constraints of any large organization. The financial services sector has been an early adopter of the technology but we see enormous potential in other sectors and are engaged in active, serious dialogue with some three dozen client opportunities outside of finance, which we expect to come to fruition in the near term.

ML. Commercial enterprises are beginning to see the benefits grid computing has to offer, much the same as the scientific industry has been experiencing for years. Grids enable efficiencies, competitive advantages and advanced research and development in ways never before imagined. Enterprises are using grid technology to allow for better customer service, provide reduced time-to-market and improved business flexibility, and to leverage existing, yet unused compute, storage and network resources to improve their fundamental business processes. Many companies are slow to adopt grids because they want proven, standards-based solutions and GGF is working hard to further develop these. Starting small and getting some experience with grids will help companies immediately experience the benefits with relatively little investment.

KK. Looking forward, customers will see more and more value from grid as standards evolve and are adopted, and as we move from number- crunching and batch oriented solutions to more transactional based grids. Furthermore, as we move towards a SOA type of application development environment, grid is going to be a key element underneath so far as helping to create a more effective and dynamic infrastructure for running applications. Over the next 3-5 years you’ll see more and more grid-based technologies and capabilities add value to helping customers be more competitive, based on having a more dynamic infrastructure.