主题：关于Linux的推广——普通用户篇 -- 万斤

I know, in many system biology labs dry-side or pure dry labs, they built their own clusters. 100k could be enough for a 128 nodes(2xDual Core) on a rack, with KVM, Centrol Terminal, even UPS. That's not a big deal. The only problem is AC.

In a supercomputer center, the nodes could be more than 1k. The reason why those labs spent money to build clusters is that, though computing hrs are free for academic users from large SCs, like SDSC, NCSA, PSC, etc, u can't get enough nodes at one time if ur application is parallel. In general, in those SCs, the Q rule is not exact FIFS, also depends on how many nodes u apply. So it's very likely for ur job to stay in the queue for more than 1 year if u want more than 100 nodes for ur program once. Actually, this market is not trivial since computers are pretty cheap so far. RO1 funding form NIH is more than 1M, it's a piece of cake to spend around 10k on computing infrastructure. However, the costs of professional clusters are not acceptable for a lab or small team, even a company, like Blue Genes.