As this issue was being published, Microsoft was getting ready to announce it's first production version of Windows Compute Cluster Server 2003 (CCS). After much anticipation, the company is finally releasing its cluster management offering. The actual announcement is a “release to manufacturing” which means the engineers at Redmond have signed off on the final version. No more betas, no more release candidates. This is it. For evaluation purposes, you should be able to download CCS from the company's website very soon, but you won't be able to purchase it until August.
CCS, designed as an extension of the Windows Server 2003, will provide a cluster management software platform for 64-bit x86 high performance computing systems. It will go up against established Linux-based solutions that now dominate HPC cluster systems worldwide.
The Linux competition will be tough, but Microsoft does have big two things in their favor: (1) Many organizations, both government and commercial, are already comfortably using Windows-based systems, everything from PCs to enterprise servers; and (2) As the largest software provider in the world, the company commands the resources to make things happen.
However, high performance computing is generally unfamiliar territory to Gates and company. Within the last year, they've managed to snag such notable HPC veterans as Tony Hey and Burton Smith, but most of Microsoft's expertise has grown up around PCs and, more recently, enterprise servers.
And because Microsoft came late to the party, rather then setting standards, the company has had to adapt to the existing HPC culture. For example, they incorporated both MPI and the Beowulf cluster model of computing into their product. In addition, they've partnered with some key cluster OEMs (HP, IBM and Dell), interconnect vendors (Myricom, Voltaire, Mellanox, and Silverstorm) and software vendors (Fluent, LSTC and MSC.Software) to make sure CCS is well supported in the HPC ecosystem.
Assuming the product generally works as expected, Microsoft should be able to pick a lot of low-hanging fruit in Windows dominated organizations that don't already have a lot of clusters deployed and are looking to scale out. CCS may also be able to compete against Linux solutions, where the end-users' workstations are Windows-based. In this environment, CCS will be able to provide a more seamless end-to-end solution.
Initially, the CCS may get the most traction in workgroup-sized clusters (less than $50K) or even smaller. In particular, it would seem to be a natural software platform for “personal supercomputers,” often defined as systems under $10K. From the user's point of view, the conceptual difference between a personal computer and a personal supercomputer is rather small, so OEMs of these machines may find the Microsoft solution very attractive. In fact, Tyan's Typhoon personal supercomputers have already been demonstrated running under CCS.
To learn more about this new product, I talked to Kyril Faenov, Microsoft's director for High Performance Computing. Our feature article outlines the product's capabilities and provides Faenov's view on how it will fit into the HPC marketplace. The folks at Redmond seem to have done their homework. Here's the money quote from Faenov:
“Ultimately, the folks that we're talking to are focused on getting their work done. They just want to be able to run a core set of applications to get their results. They really don't particularly care what the underlying system is.”
It'll be interesting to see how this unfolds over the next year.
Also in This Issue — a Sterling Interview
If clusters are not your passion, catch the Thomas Sterling interview this week. Sterling, along with Don Becker, developed the original Beowulf cluster computing model, but now confines his interests mainly to high-end supercomputing technologies. Contributing editor, Chris Lazou caught up with him at the NEC User Group meeting last month in Toronto. The resulting interview is one of the best in HPCwire this year.
Lazou and Sterling discussed many interesting topics: silicon technologies, optical computing, federal funding of HPC, DARPA's HPCS program, Petaflop computing and Sterling's own research work. I thought the most interesting exchange was when Lazou brought up the need for new HPC languages and the chicken-and-egg problem of getting software vendors to support them. Here's what Sterling had to say:
“Every language was a new language at one time. Most are not accepted by the mainstream. Some are adopted for niche applications. Very few have a major impact. The thing to remember is that over the next sixty years (the length of all electronic computing to date), we will write many times as much software than already exists today. The true legacy code is our future not our past.”
I hope someone at the DARPA HPCS office is listening.
As always, comments about HPCwire are welcomed and encouraged. Write to me, Michael Feldman, at [email protected].