BM Seer Unofficial thoughts from an anonymous Sun employee

Power efficiency means getting denominator right

Tuesday Mar 20, 2007

Lots of people are talking about getting the datacenter power down and measuring it in the right way. But the metric needs to be right.

The best way to build the proper metric is to base it on previously proven metrics. Price/Performance ($/perf) is a fine metric to look at. Because most datacenter people (and their finance people) are worried about watts because it is a cost to the business. Actually in the final analysis I'm more worried about it because of the environment.

Both of these metrics are a 'lower is better' metric, so the marketing people who like big numbers may not like it, but they just like bravado. What businesses want is a smaller number here (less cost, less watts)

Price/performance
watt/performance
...it just makes sense. And it follows that when you look at datacenters and add in watts to cool this metric also shows the right things.

So tell anyone interested in gauging the power efficiency to please use this metric. I'll admit that I've probably used performance/watt myself, but that's what happens when you rush things.

More on this tomorrow its getting late...

Like this post? del.icio.us | furl | slashdot | technorati | digg

Full Rack Woodcrest? 10.2KW or more?

Tuesday Oct 31, 2006

Open question: Anyone have a full rack of 2-socket 32GB woodcrest servers yet? If so what is your total power draw? From the posting earlier today we've found a public reference of 510watt/2-socket woodcrest server with 32GB of memory. That is 510*20 or 10.2KW per rack. ...but actual rack measurements would be appreciated.

If you had more disk, like what is shipped with it I'm guess more like 12KW? Also how much whitespace do you need in a datacenter?

[1] Comments
Like this post? del.icio.us | furl | slashdot | technorati | digg