slacker (s@lack.er) on 4/10/09 wrote:
---------------------------
>Jack (Jumpingjack6@verizon.net) on 4/10/09 wrote:
>---------------------------
>>Peak power will dictate various thermal cooling designs and such, but total energy
>>consumed determines energy efficiency. A server can have 20% higher peak power
>>but if it completes the task in 50% less time will be more energy efficient.
>
>Only guaranteed to be true if you're strictly looking at the time span to complete
>the task on each server. Otherwise, you need to add the idle energy consumption
>when the faster server is sitting idle. This is what shot the Harpertown platform
>in the foot for workloads which you can imagine.
>
>David: consider writing a short article explaining energy vs. power, and where each metric is relevant.

True, I simplified it. The total energy over a long period depends on many factors, in terms of energy savings the sooner a task completes and gets to idle, caveat -- idle needs to be as low or lower than the competing or existing solution to be of benefit.

The concept is straight forward. A box that demonstrates peak power that is higher can actually be more energy efficient.

jack