Miller –
The world’s first Commercial High Performance Computing Conference and Expo was held in a pink, yellow and blue Hyatt Hotel in Orlando, Florida, this week. I spent a few hours there because I figured — correctly — that Linux would be one of the major discussion topics.
This was a gathering of people interested in or selling industrial-grade cluster and grid computing hardware and software, with presentation titles like, “Managing IT on a Global Scale, A Case Study on the World’s Largest Information Technology Company.” This was not a sales pitch by IBM. It was about their own, internal operations, and the audience consisted of people from Sun, Intel, and a number of end user (non-technology) companies that are starting to explore grid computing and wanted to learn from IBM’s experience.
There was not much of an exhibition floor; a few booths in a small room was the extent of it. But it was a Linux-heavy exhibition floor, with IBM’s demo clusters booting Red Hat Linux, and Scyld not only there but quite prominent, along with Linux NetworX and several less well-known companies selling Linux-based, enterprise-level hardware and software. Intel was there — with Scyld logos visible in its booth. And Sun was there in force, the conference’s largest sponsor, touting both Solaris and Linux products.
Sun has been saying “the network is the computer” for a long time, and that’s what this grid computing stuff is all about.
Grid computing is not a new idea
Rajkumar Buyya, of Monash University, says he has been working with and writing about grid computing since 1990 or so, “except we didn’t call it grid computing back then. We used to call it networked computing or distributed computing. Grid computing is the same thing with a catchier name.”
To Buyya, the biggest recent advance in distributed computing — aside from the new, catchy name — is its move from academic curiosity out into the world of private industry. Buyya’s presentation was titled, “Weaving the World Wide Grid Marketplace: Economic Paradigm for Distributed Resource Management and Scheduling for Grid Computing,” and his presence at this conference was sponsored by GridFrastructure, a Massachussetts company that says it is “dedicated to supporting the deployment of the global grid,” and seems to have staked its entire future on distributed high performance computing.
When you think about it, most of this grid computing stuff is an extension of the SETI@home project, jazzed and prettied up, either being run internally by companies like IBM or sold as a service by GridFrastructure and others. And, according to Patrick Dreher, associate director of MIT’s Laboratory for Nuclear Science, the move toward more grid computing is inevitable. “We’re dealing with more data every year, he says. “From terabytes we’re moving to petabytes.” And Dreher points out that bandwidth is becoming “practically free” so that moving all this data from one clustered supercomputing node to another is no longer as big a deal as it once was.
Dreher’s grid computing vision is not the old SETI@home idea of harnessing the power of individual home computers, but is based on clusters that are becoming more powerful and less costly, and connecting to underutilized clusters elsewhere if and when they are confronted with a calculation set beyond their ability. Dreher sees his cluster connecting “first with ones in other departments here, then with ones at other universities.” In other words, creating Beowulf clusters of Beowulf clusters on an ad hoc basis instead of each company or university or (as is common today) each lab or department within each company or university trying to own all the computing capacity it might possibly, conceivably ever need.
The biochem and pharmaceutical industries were mentioned over and over as some of the biggest potential users of grid computing, and apparently the biotech people agree — and are also gung-ho on Linux and Open Source tools. But there is a thrust toward bringing advanced clustering and grid computing techniques to other businesses that may not need them constantly but only once in a while. This is where grid computing may eventually find its best use, allowing companies or even individuals who want to do a particularly complicated rendering or create a complex model of some sort to access a global network of networked computers for a few minutes or hours for comparatively few dollars.
More about the conference itself
Elaine Mershon, an employee of conference organizer INT Media Group told me about 400 people had preregistered and that about 200 had actually showed up. This dropoff in tech conference attendance has become typical in the last year with the tech industry recession, and combined with many people’s reluctance to fly since last September. I was only around for part of the conference’s second day, but the one time I tried to take a head count in all three occupied conference rooms, I found only about 110 people, including those hanging around in the lobby and a few lurking on the small exhibit floor.
Later I overheard Elaine and some of her coworkers talking about a severe second-day attendance drop. But there were still some interesting people floating around, and making casual contact is often the most important aspect of this kind of conference. For instance, I accidentally wandered into a conversation that included an Intel senior sysadmin talking about how his company is planning to switch all of its 26,000 Unix servers and 17,000 Windows NT/2000 servers over to Linux over the next few years to create a single, seamless, worldwide operational grid. A conversation over coffee with Etnus CEO Chris Doehlert became a discussion of software licensing issues and how to reconcile Open Source and Free Software philosophies with the need to earn a living, a conundrum Chris wrestles with — and we hope he’ll choose to write about it for us at some point. (Yes, I asked.)
The guys at the Linux NetworX booth wanted to know where the NewsVac summaries on the NewsForge front page had gone, and I told them the same thing I’ve told everyone else — they’ll be back as soon as we solve some code issues with the new NewsForge, so relax and be patient. The show was disappointing for Linux NetworX as a sales venue. They said they had hardly any real sales leads but, as one of their sales engineers pointed out, “It’s the first show like this, ever. The next one will have more people, and we’ll almost certainly be there.”
Commercial grid computing is just beginning
Even though this was billed as a commercial high performance computing conference and expo, a high percentage of attendees were connected with universities and research institutions.
The biggest grid computing dreams I heard discussed were those revolving around computing services becoming a metered, commercial utility like electricity is today, where users just plug into the grid and neither know nor care where the power they are using comes from. These are nowhere near reality. But every IT industry advance starts as a dream, and dreamers must gather and share ideas with each other, especially when their dreams involve worldwide linkages, not standalone computers.
Right now, grid computing is used either by research or academic institutions alone or in groups, and by commercial companies almost strictly as an internal, non-shared way to gain increased computing efficiency. Everything beyond that, as far as commercial grid computing, is still being worked out, and there is a lot to work out, including interconnection standards, security, and how to charge for computing services in a fair (and marketable) manner.
That’s the biggest challenge of all to market-oriented distributed computing people: how to provide large-scale, utility-style grid computing service at a profit.
Remember Buyya, and his presentation, “Weaving the World Wide Grid Marketplace: Economic Paradigm for Distributed Resource Management and Scheduling for Grid Computing?” Buyya has spent years working on ways to sell networked computing service, but when I asked him if anyone had actually managed to make money providing utility-style grid computer service yet, he answered with one word:
“No.”
