I read this article about the state of windows machines in the top500 list of “fastest” supercomputers. Remember that Microsoft indicates that it has no interest in the top500, and given its purported strategy, that sounds like it is correct, that they shouldn’t care about “non-mainstream” supercomputers.
Since Microsoft appears to want to make supercomputers appear to be simply big PCs that are out of sight, focusing on top500 doesn’t make much sense.
Regardless, one of the NCSA machines was running windows. For at least a portion of its existence. When it was upgraded, they seem to have switched to Linux. Or at least booted into Linux.
When the November Top 500 list was made public earlier this month, Windows Compute Cluster Server 2003 was nowhere to be found. As it turns out, Lincoln is a dual-boot machine, and the NCSA decided to give Linux a run on a beefed-up Lincoln. When it was all said and done, the Dell cluster of 2,048 processors had scored a sustained capacity of 16.5 teraflops and a peak performance of 21.3 teraflops, which gave it a ranking of 27 on the Top 500 list. The other Windows CCS machine was dropped from the November list; a system needed to deliver 2.7 teraflops of sustained capacity to make it into the top 500.
What would be interesting to me, and others, is the relative performance of this machine running each OS. That is, define delta to be score under windows – score under linux for the peak and Rmax tests. Positive deltas would be reported, and negative ones would probably be covered up. I would expect a delta near zero in either case if the Intel/PGI compilers were used for both OSes. I would expect a strong negative score if the Microsoft compilers were used (not that they are “bad” just that they may not optimize for HPC calculations as well as the Intel/PGI compilers).
The subsequent paragraph to the quoted one is, well, amusing. Just goes to show that marketeers are everywhere, and high angular momentum states (e.g. lots of spin) occur in things that are perceived to be of real marketing value.
I don’t think that Microsoft should aim for that list. Many reasons why not, but the important one is that this is not where they claim their market to be. I don’t expect windows to be on the worlds fastest/largest machines. There aren’t many such machines, and Microsoft is a volume player. If it does score a slot, I have doubts that the machine would remain a windows machine unless Microsoft paid for the machine and required the facility to not run any other OS on it in order to receive it. But that sounds like a bad deal to me, all around. Bad marketing direction, bad overall idea.
I still claim that Microsoft’s greatest value in HPC is in the potential for interoperability, making it really easy to hook into supercomputers. Not windows clusters, but general supercomputers, which include Linux, Unix, and a few windows machines. I don’t think there is great value to their customer base in building larger collections of windows machines. Their customers must not agree that there is great value in this either, as they appear to quickly convert such systems to non-windows machines.