Friday, September 11, 2009

Ultra-low PUE??? KC Mares seems to know so.

KC Mares of MegaWatt Consulting has managed to get some really low PUE with today's technology. Color me sceptical but KC says his math is holding up up.

Now, you ask, how did we get to a PUE of 1.05? Let me hopefully answer a few of your questions: 1) yes, based on annual hourly site weather data; 2) all three have densities of 400-500 watts/sf; 3) all three are roughly Tier III to Tier III+, so all have roughly N+1 (I explain a little more below); 4) all three are in climates that exceed 90F in summer; 5) none use a body of water to transfer heat (i.e. lake, river, etc); 6) all are roughly 10 MWs of IT load, so pretty normal size; 7) all operate within TC9.9 recommended ranges except for a few hours a year within the  allowable range; and most importantly, 8) all have construction budgets equal to or LESS than standard data center construction. Oh, and one more thing: even though each of these sites have some renewable energy generation, this is not counted in the PUE to reduce it; I don’t believe that is in the spirit of the metric.
Google has reported 1.2 and 1.10 but if KC is right then they could possibly do even better. That all said I look forward to seeing if the test of time bears it out. That is one problem I have with PUE. At this point it is all theory and short term testing. At least as far as I have seen.

The question "Did all the IT load really NEED to be on UPS? " has some very interesting ideas. Yes, it comes down to risk but it is a very serious question that should be asked. In most cases the UPS is simply there to carry your load long enough to transfer power to the generator. Well why do you need both of your power supplies on the UPS for the preparation of a 15 minute power outage? Why have two when one would carry you through that time?

Of course you could point out "well what if your power supply on a critical server fails while your load is being transfered?" To which you must ask, "how 'critical' is this system and if it is that super critical why is it not clustered and have a failover server as well? Or do you like your single points of failure on one piece of hardware?"

I shall ponder this more. As they say the more direct you can get your power to the equipment the less power you lose. That cuts out one big middle man. Not sure APC would be all that happy...