Microsoft sees HPC for non-geeks

AFTER ALL THESE years, it may be that high-performance computing (HPC) is ready to lose its tag of being for a chosen few.

The prime mover here is Microsoft which last week sent final code for its Windows Compute Cluster Server (CCS) 2003 operating system to manufacturing. That's a complex name for a complex product but Microsoft's plan focuses on simplicity. It wants to make more of us applicable for a scale of computing that would quite recently have cost seven-figure sums but can now be had for tens of thousands of pounds.

Microsoft's pitch makes practical sense. It wants to take advantage of the skyrocketing number of Mips that can be had for a increasingly modest outlay to put a supercomputer "under the desk" rather than locked away to be used only by granted privilege.

With multicore processors helping to prop up Moore's Law and a similar bang-for-buck equation applying to storage, supercomputing, however we define it today, is being democratised even without Microsoft's intervention. Of course, this curve has been described for decades and together with the necessary cluster networking standards and protocols make it a no-brainer to adopt scale-out computing rather than scaling up for most applications.

Also, 64-bit addressing has of course become a standard feature of the most popular microprocessors, helping to make today's commodity chips applicable for large data sets characteristic of HPC. Microsoft's specification of x64 Opteron and Xeon processors for CCS may well be remembered as one the nails in the Itanium coffin.

As well as riding the hardware pricing wave, Microsoft sees itself as having a differentiator. The plan revolves around supplying a set of user and admin tools that capitalise on the firm's familiar look and feel so that clusters aren't aimed at Linux and Unix pointy-heads but can be used with relative familiarity by experts in their own academic, medical and commercial domains. Universities are an obvious target audience but finance firms' Excel-addicted risk analysis routines, digital-video rendering and other fields are obvious candidates.

Microsoft has a lot going for it. Rock-solid relationships with the IHV community mean that IBM, HP and Dell will be ready in August with pre-configured boxes and its undoubted nous at recruiting ISVs is also likely to pay dividends. It has also done plenty of seeding with academia, recruiting a key manager from Southampton University, for example, and sponsoring programmes at various colleges. Those pre-existing relationships can't harm Microsoft's chances of bagging more academic and research buyers.

What is yet to be proven is how much of a need there is for HPC beyond its current reach and to what extent Microsoft usability will be favoured over Linux/Unix, the current platform of choice for HPC. Performance characteristics are also yet to be established.

Microsoft has taken another step into new ground for Windows but it remains to be seen whether Windows for supercomputers can do what Windows NT did on servers or whether it will struggle like the company's efforts in handheld devices. µ