With the release of Windows HPC Server 2008, Microsoft is attempting to make up for its late entry into the high performance computing market. Whereas its first generation product, Windows Compute Cluster Server 2003, got its foot in the supercomputing door in 2006, Microsoft is hoping its 2008 offering will give Linux a run for its money.
The HPC group at Microsoft isn’t looking to unseat Linux on current systems. Rather, they’re counting on expanding with the overall HPC market. Since a lot of this growth is expected to take place at the workgroup and department levels on x86-based machines, Microsoft is in a good position to pick up new systems as users move from Windows workstations to small clusters and from small clusters to mid-sized systems.
The engineers at Redmond have been busy adding features to their HPC platform over the last couple years. The big talking points are the new high-speed NetworkDirect RDMA stack, a service-oriented architecture (SOA) job scheduler, and enhanced cluster interoperability via the Open Grid Forum’s High Performance Computing Basic Profile standard. The overall strategy here is to offer a full-featured software stack on top of an OS, rather than leaving it up to customers or third party integrators to forage for software on their own.
Since Linux vendors like Red Hat and Novell don’t seem to be that interested in building integrated HPC products, this gave Microsoft an opening to develop a more turnkey solution. To be fair, last year Red Hat and Platform Computing teamed up to offer an integrated Linux HPC platform — Red Hat HPC. But that offering just bundles Platform’s Open Cluster Stack with Red Hat Enterprise Linux. I’m not sure how much time or money that saves than if those pieces were gathered and configured separately. But one thing you do get with these integrated solutions is certification on various hardware platforms. For commercial users especially, this can be just as important as the feature list.
Since I don’t have a cluster in my basement to play around with, I’m not qualified to judge the relative strengths of the Windows HPC platform versus a Linux-based one (If anyone out there has done such a comparison, I’d be interested to hear from you.) But one thing Microsoft has done very well since auditioning its first HPC product two years ago is built up an HPC vendor ecosystem. The company has managed to line up some of the biggest OEMs, ISVs, storage vendors, component makers and system integrators in the community.
Currently Microsoft lists more than 60 HPC partners. On the software side, everyone from ANSYS to Wolfram Research has hopped aboard. The range of OEMs that will be offering the Windows HPC server is particularly impressive. The three biggies — IBM, HP and Dell — have signed up. IBM is even offering $99 test drives on the Windows HPC server via its Computing On Demand setup. With the recent announcement of Cray’s CX1 mini-super, even the venerated supercomputer maker is now in the Microsoft column. To cover international real estate, Microsoft has latched onto Dawning, Bull and NEC. Those OEM partnerships mean that Microsoft now has access to about 85 percent of the x86 portion of the HPC server market.
A handful of HPC system vendors are still holding out. These include Sun Microsystems, SGI, Appro, Penguin Computing and Rackable Systems. But if Microsoft manages to grow its footprint in the HPC space with its new server platform, I imagine all of these vendors will be making the trip to Redmond.