INDUSTRY

By Steve Fisher, Editor In Chief -- Platform, Compaq and IDC hosted a Web seminar or Webinar last week entitled, ‘Grid Computing: Beyond the Hype.’ It was not what I would consider truly “hype-free” as I heard more than a few HPC industry buzz words thrown around. Granted, they were some of the newer ones which was nice, but they are what they are. Then again, I’m probably being too hard on the organizers as I am nearly convinced that we live in a world where a “hype-free” environment may not exist. Anyway, the “Webinar” was quite informative and featured the remarks of some VERY knowledgeable and respected figures in HPC: IDC’s Debra Goldfarb, Ty Rabe of Compaq and Platform’s Ian Baird. Goldfarb’s presentation was thorough and informative and Rabe and Baird’s answers to audience questions were well spoken and insightful. I was reckoning that part of the reason the event was planned was the intention of delivering “Grid Computing” from the ranks of industry terms such as “Total Cost of Ownership” whose definitions remain nebulous at best. Debra Goldfarb delivered some “straight talk” right away that I really liked. “I think the bottom line here is, don’t get hung-up on definitions. One person’s cluster could be another person’s grid. It really depends on how you look at the market.” She is right, but I feel the need to “define” so I have to go against her advice. What is a grid? The following is the definition I walked away from the Webinar with. Grids are: an emerging infrastructure that enables the integrated use of basically any type of device whether it’s a high end computer such as a supercomputer, databases, instrumentation devices, networks etc. This infrastructure allows for the distribution of workloads, the ability to tackle highly complex problems and to potentially provide a pool of computing resources which fall under a utility class of model. The Webinar also highlighted the following types of grids: Compute grids: virtualized pools of compute resources Data grids: Grids with shared access to databases and files. Utility grids: Grids where you’re basically sharing access to applications, software, and computing resources. A broader dimension of the grid type. None of these is unique in and of itself. They share certain attributes and they overlap in certain dimensions, but they (the three different varieties of grids described above) viewed together under the grid “model” are more comprehensive than what most would call distributed computing systems. I think that’s one key differentiator between the “grid” and “distributed computing” buzz words/phrases. I’ll leave the final analysis of differences between the two models to better minds. I will say this though, the concept of distributed architectures is continuing to develop, which in my opinion is a very good thing. Now, that being said, one could say, “Hey, I’m super glad they’re developing, but can we use grids today?” “Grids are really being used and are in place today with early adopters and are seen in a lot of research communities and at a few major corporations that have multiple sites…” said Compaq’s Rabe. “So in terms of deployment, there are a relatively small fraction of people that have them deployed. In terms of where we’re all headed I think we’re all confident that they’re going to be rapidly deployed in a growing number of locations over the next several years. So we’re keenly interested in watching that evolution.” So it seems that public grids with access for us consumer-types are a still a thing to come. One reason behind that may be security. If you think about it, there are a lot of holes that need to be plugged when undertaking something on this grand a scale. However, there are organizations like the Global Grid Forum that are hard at work developing technologies and architectures to address security issues that will follow standards around encryption, authentication, authorization…the standard forms of security as we use them today as they’re adapted to grids. Widespread grid access for corporate and research people will very likely be along much sooner. Platform’s Baird commented, “As grids are using IP technology, then the obvious questions that come up around Internet security are there. So one of the areas that we believe grids will take on their quickest acceptance and early adoption will be in enterprise grids where all of the sharing of resources etc. are grids within an organization and thereby within the firewall of those organizations.” Another issue in grid computing adoption will be bandwidth. Compaq’s Rabe seems to think things are in pretty decent shape now and will only get better over time. “It’s very clear that bandwidth of the network or whatever kind of connection is being used will be a constraining factor in the growth of grids. So we are seeing dramatic increases in the bandwidth of networking technology and so the idea of providing access across many remote sites to specialized resources is real today and will become increasingly popular.” Baird agreed, “Some of the things too that are applicable to grids can usually be parsed into very small packets of information and be distributed fairly quickly over the infrastructure that exists today. And furthermore there’s the bandwidth that is developing today which is growing at an exponential rate. Such a rate that it won’t become the bottleneck that it has been in the past, in the future. And finally when you talk about data localities, grids in themselves and the applications that will manage the grids will help to manage data and its ability to know where data is and to effectively have virtual data so that you do not have to deal with the issues around data locality as much as we do today.” Touching on HPC issues Rabe said, “I think where there may be a misconception is in the case of the most demanding high performance computing problems. There really are specialized requirements for interconnects in terms of very low latency and very high bandwidth. Those kinds of applications won’t be running on the grid any time soon, but the grid will enable access to supercomputers that can run those kind of jobs and make that access available to a much broader set of users.” In conclusion, I found the ‘Grid Computing: Beyond the Hype’ Web seminar (Webinar) to be very informative and delivered in an interesting format. It reminded me that grids are an evolutionary concept that have been brewing and growing over the last ten years or so. It also hammered home the fact that in the minds of a great many people, grids are the future of computing for our planet. A networked world far beyond what we have today. I’ve heard the term “knowledge grid” once or twice and I like it. I think that’s something we can all look forward to.