Ultra-green Europeans scorn Facebook's data centre blueprint

McDonald's refugee

On the basis of that quote it does kind of make you wonder how he escaped working at a McDonalds.

I've been reading alot about NASA computing and something called STAR which was a fault tolerant system for space probes. basically the idea was to have 3 or more of all components - two of which were unpowered backups - thus increasing the mtbf which in probes that are a little out of AA or RAC jurisdiction is useful.

Why not adapt the STAR idea for data centres creating systems that have multiple multicore processors - the majority of which are only powered at high usage rates. If a current system uses say two 12 core processors at 65% but powers all its cores its wasting power compared to a system that uses say 6 x 4 core equivalents of which one or more is only prefetched and run when load requires. This would give the added advantage that if a processor starts to throw errors it can be demoted from a primary to a load dependent one.

It makes the software and the initial hardware outlay a little more complex/expensive - but means savings in downtime & cooling etc, not to mention power bills..