Vendors aim to put a cluster in every department.
SGI and Microsoft announced today that they had partnered on a cluster solution aimed at the middle-market. SGI says the alliance addresses their customers in scientific fields that have Microsoft technical workstations and are looking for tighter integration with their compute cluster or possibly don’t have Linux administration experience. The solution will run on SGI’s Intel Xeon-based Altix XE servers with Windows Computer Cluster Server 2003 as the OS.
What’s most interesting is the way in which they are talking about the solution. Not so much in engineering terms but in language that would make sense to just about anyone in an enterprise setting.
“We want users to be able to select compute resources in the same way that they select a print resource,” said Shawn Hansen, director of HPC marketing at Microsoft. “Just as you would say, ‘print this document to the 4th floor printer,’ you can say, ‘run this job on the 4th floor cluster.’ No direct access to the cluster is needed, you could run it from your laptop and the cluster would know you via standard Windows credentials.”
Commodity cluster vendors like to talk about printers because their goal often attempts to mirror printer distribution. Just as every department has it’s own printer, it should have their own cluster, it is sometimes said. However, as SGI is probably already aware — having migrated from high-end Irix systems to Linux-based commodity clusters in recent years — mid-market HPC is a tough nut to crack.
Cluster vendors have been trying to make inroads to the general enterprise market for almost as long as there has been a vendor ecosystem around HPC. The reason for this is pretty simple: there are only so many National Lab contracts to go around and, as it stands, way too many vendors selling extremely similar technology solutions.
Because of this, the commodity cluster market is poised to go the path of the commodity PC market: aggressive and rather painful consolidation. I say aggressive because my feeling is that the venture capital that has supported this market the past several years likely won’t last beyond 2007.
And I say painful, because the acquisitions of unique technology (think blades and networking) have already been made. If you acquire an HPC company now you would probably get a lot of engineering talent but you wouldn’t have to pay the premium you would if you were also buying a highly profitable product line.
It’s not that cluster sales overall are terrible — many original equipment vendors seem to be doing quite well — it’s just that they are being dominated by a handful of solutions vendors.
If Linux dominates clusters then the lack of mass market adoption can be attributed to one of two things: mid-size enterprise doesn’t need HPC or Linux just isn’t a fit.
Microsoft thinks they see some opportunity in the latter. Enterprise HPC may very well be an education issue — corporations may not know they need HPC yet — and toward that end, MS has build distributed computing into Excel 12. However, I suspect even with Windows-friendly customers, they’ve quite an uphill battle ahead of them.