wattage charts to think about
Friday Nov 17, 2006
Just got back from dinner with a friend who works for Intel. He drew me a graph that looks like this:
Problem is that this kind of graph is very misleading. Clearly I need to by him a copy of Edward Tufte's "The Visual Display of Quantitative Information" for Christmas. Tufte has a nice section on graphing. Because his Intel-inspired graph gives you the impression that watts fall exactly with utilization. Here is a truer picture, as you need to graph watts from 0 or total watts, just like you graphed 0% to 100%.
Lesson is the look at the charts to make sure someone isn't making changes look bigger than they are.
But actually as I showed earlier today in a graph of a 300 watt server at 100% that falls to 120 watts at 10% or idle, the real wattage used to do the work goes up hyperbolically. Again this is comparison to 100%, now I know that you won't be doing it at 100% but you can see the optimising for low utilisation is the wrong thing to do.
Remember from a posting earlier today, a better thought experiment gives more info. Let's compare 5 servers running at 10% utilisation (that is 120 watts each or 600 watts for the 5 of them). A single server running at 50% utilisation only uses 200 watts, therefore the 10% case almost require 3 times more power!
|
|
|
|
