Nicholas Carr has two posts on this. In the first he confirms that Rackable’s solution is, in fact, prettier. In the second he adds some analysis and points to two points of view as articulated by Sun’s Greg Papadopoulos and Microsoft’s James Hamilton. Nick’s post includes links to papers and presentations by both (I don’t actually know him, but I’ve decided that I can call him “Nick.”)
At the high end of HPC (say, the top 150 of the Top500) I don’t think this matters very much. The big centers with stable funding have benefitted enormously from the dramatic contraction in the size of computers over the past decade. The obvious exception here is the top 2 or 3 slots at the lunatic fringe of big. These guys pave over counties to house each new machine they buy.
Where containerized computing has the most potential to matter is for people who don’t have any kind of HPC or large computing experience. They like don’t have legacy floorspace they can recapture. And it’s this very market that we need to do everything we can do to encourage entry into HPC. Disruptive discoveries will be made by new people and new disciplines using HPC for the first time.
Dan Ciruli brings it all home with a reminder that no one in this market can use HPC right now because it’s too hard (I’ll be talking about this, and why it matters even at the high end, at HPCC next week and will post my talk here for those interested).
This is an interesting market experiment with two questions: can the companies execute, and even if they can, is there a market?