You can answer a question for me on Linux then, (more of a, "Did you ever?") were you privy to making the clusters from PS3 units? I was deeply intrigued with them, but I'm not quite funded for the endeavor in the classical sense.
P.s., welcome! I have a poker II with the browns, it's worlds above rubber dome drudgery.
I understand why someone might want to build a cluster of PS3 consoles, since Sony supposedly sells the console at a loss to bolster the user base. I have not and would not build a cluster from PS3 consoles. Three big reasons for this...
One, the PS3 is not meant for a data center. You couldn't rack and stack them. You'd have to put them on shelves, which is bad for density, airflow and cable management. They have no baseboard management, so there would be no niceties like remote console or automated install options. You would have to deploy smart PDUs, KVM multiplexers, monitoring systems, etc to bandaid the management situation. Do not want.
Two, the config does not have a wide deployment or enterprise hardware and OS support. I don't want to be the guy all by himself, running into unsolvable and unique problems. I buy major vendor hardware that supports Enterprise Linux because that hardware and OS has been deployed together and tested in production through multiple hardware revisions. I get to talk to systems engineers that have the experience of solving problems across many hundred, thousands of similar deployments.
Three, the Cell processor is hard to program efficiently. Most researchers do not want to rewrite all of their code for a specific hardware architecture. They don't mind using a better compiler, but even a fancy compiler can't magically make your algorithm go parallel well. Programming for a Cell is like programming for a co-processor; It is not as simple as having more cores and adding a compiler pragma.
I hope that answered your question!
