Re: Performance testing

Josh:

Predicting performance is a tricky business, precisely because there are so many variables. CPU speed might not even be the dominant factor.

For example, you did not mention how much real main storage is installed on the 520 or the new 720, or how much DASD (total GB, and total number of disk units) on each. The number of disk units matters because it directly relates to the number of disk heads and arms available. And, what kind of disk controllers are involved, such as RAID controllers, with how much cache memory, etc. -- all of these factors can influence the overall performance of your applications. You might also need to consider how the subsystems are configured, how memory pools are allocated to those subsystems, etc.

Since you are "running the boxes in parallel" then why not just run the real jobs on the new box, especially any long running batch jobs, and see how long they take? Then you can directly compare "wall clock time" between the old and new systems, for your workload. This would give you the best and most accurate idea of what to expect.

Hope that helps,

Mark S. Waterbury

> On 12/17/2010 9:22 AM, Josh Diggs wrote:

We are actually running the boxes in parrallel now and still in production on the old one. We could pull performance numbers off of the old box, but we won't know how that compares until we cut over to the new box. I was looking for more of a predictive analysis that would let me reassure management that we were going to see some performance improvement.

Thanks for all the great ideas.