Re: Performance modeling

From: "rob"
Is it reasonable to estimate based on CPW when you aren't using
the entire box for the one lpar? Here I have 4 lpars on production and
5 on the HA.

It seems to me that CPW would be the best metric to base estimates on.

Some years ago IBM also started publishing the CIW (compute intensive workload) metric in the System i Handbook, but there always seemed to be a strong correlation between CPW and CIW. It seems that regardless of the claims that CPW is based on memory and disks, published CIW numbers seem to indicate that CPW is more dependent on CPU speed, and IBM's former manipulations to throttle CPU on entry-level servers.

I've never paid much attention to lpar technology, but I assume there is some overhead associated with each one allocated. I don't know what that would be. And of course, lpar performance would normally depend on the number of cores allocated to it.

Standard disclaimers apply to topics like this. Your mileage may vary.

-Nathan