Take a Windows HPC test drive
rashing in on IBM's supercomputers
With Windows HPC Server 2008, the parallel supercomputing variant of Microsoft's Windows operating system, being released this week, there are probably a lot more people who want to see how it works than are willing to shell out a lot of cash to give it a whirl. And that is one reason why IBM has set up a test drive of Windows HPC Server on a number of compute utilities.
The other reason is Big Blue is doing this is to remind us all, once again, that it has compute utilities in the first place, since the utility computing model - I guess I meant to say cloud computing to be more modern, but somehow forgot - has not taken off as many (press releases) had expected.
IBM operates two HPC utilities, one in New York and the other in London, that together have a mix of over 13,000 Power and x64 processors (both from Advanced Micro Devices and Intel, and in blade or rack formats) that until now have been configured with AIX (on Power) or Linux (on Power or x64); the servers in the centers have access to a total of 54 terabytes of disk capacity.
Because different customers have different networking preferences, IBM can link server nodes with Gigabit Ethernet, InfiniBand, or Myrinet interconnections, and the storage can optionally be equipped with IBM's own General Parallel File System. The company can also bring BlueGene/L supercomputer nodes to bear if users want to buy capacity on these boxes.
Starting this week, the x64 boxes can also now be configured with Windows HPC Server 2008, and rather than making customers sign a relatively long term contract, the company is offering a three-day trial across from 14 to 16 x64 server nodes running the Windows stack for a mere $99.
Based on the $1 per CPU per hour charges that Sun has tried to get for its Solaris utility, IBM's pricing for the Windows HPC test drive is very reasonable. Amazon's EC2 grid charges a mere 10 cents per hour per compute unit, which is roughly equivalent to a single-core x64 processor running at around 1 GHz, and Amazon's pricing is not directly applicable to IBM's test drive (if you make some assumptions about core counts and clock speeds, IBM is in the range of a few pennies per 1 GHz per hour on the x64 iron). What is clear is that Amazon is keeping prices pretty low for EC2, and IBM is doing so with Windows HPC Server to test demand.
Heaven only knows what IBM really charges for utility computing once customers make a commitment. Utility computing deals are all negotiated individually, and list price is probably even less meaningful than it is on a hardware sale proper. ®