all this, and Windows too!
Rather than post some vicious remarks about Windows, and its current crop of problems, I simply wonder why Microsoft is working on this.
The dollars involved must be massive to produce, and the rewards minimal, because the people who operate supercomputers are already familiar with either Unix or Linux, and know quite well how to get things done on those operating systems.
Linux clustering software is already far ahead of any Microsoft efforts, so again, the question must be, "Where is the reward?" Microsoft is more money driven than most companies, so where is the benefit to this?
The pocket protector and beanie crowd are not going to give much notice to this, until something earth shattering is developed, and then, small possible market.
One month ahead of the next round of Top 500 supercomputer rankings, Microsoft has released online its Beta 2 for its high-performance data center version of Windows. But is Windows accelerating fast enough to keep up with Linux clusters?
Windows has historically not been considered the fastest operating system among supercomputers, but there is no doubt it’s working to climb the ladder. Microsoft’s best performing trial supercomputer to date has been given the nickname "Rainier." It’s a cluster of Dell PowerEdge 1855 blades, which are currently running Microsoft’s latest test builds of HPC Server 2008 — its successor to Compute Cluster Server 2003.
In June of 2007, Rainier placed #106 on the University of Mannheim’s Top 500 supercomputers list, with a peak performance (Rmax) ranking of just under 9 teraflops. In just five months’ time, Microsoft’s later builds helped the same system accelerate to 11.75 teraflops, though that boost was pretty much in line with its competition: it placed #116 on the November 2007 list.
The big improvement Microsoft is trying to build — maybe as important, at least, as being faster — is making feasible an easily administered system. Taking some cues from its latest improvements to the admin console for Windows Server 2008, Microsoft is adding an Outlook-like front end to HPC Server, which unlike almost any other Microsoft product is apparently keeping its nickname: "Cluster Rocket" will serve as the main admin control panel, giving data cluster admins access to tools such as temperature maps, network configuration maps, and failover clustering itineraries.
And for anyone — especially in academia — who preferred Linux because the command line seemed more "native" somehow, HPC Server throws in PowerShell — the tool that reduces colossal scripts to two or three lines of code.
Truthfully, I don’t think that PowerShell is going to make any Fast Fourier analyses any easier. Those who know this type of computing are typically using matrix manipulation, so the commands of PowerShell just wont do. Something like Rexx, which was included by IBM with OS/2 would probably be more helpful.
But the real tale of the tape for HPC Server could come next month, when Mannheim U. reveals its latest semi-annual Top 500 results. Microsoft appears to be setting a benchmark for itself for 30% performance improvement with each new list, which would help it keep pace with the rest of the industry. Any less than that amount, and academic customers could fail to be impressed.
And then Mr. Ballmer would have to do another rendition of ‘Dance Monkeyboy’. (Sorry, I just couldn’t help it)
[tags] Microsoft, supercomputing, Cray computers, Linux clustering, Windows for Supercomputers beta, Dance Monkeyboy [/tags]