Monday, May 10, 2010

Benchmarking best practices

Along with having the right tools, you also need to know what you are hoping to achieve. The makeup of your workload is critical:
# IO/sec
% read vs % write
avg read size
avg write size
% random vs sequential
% cache hit

many of these numbers can be identified from a live environment using tools such as perfmon or San tools. For new environments, some creative estimating is needed.

Once you have your estimated workload, you can begin testing against the live environment. It is critical to monitor this testing inultuple locations: computer, fc network, cache, and disk access. All these monitors should tell the same performance story. If they don't, look for caching, block size changes, filesystem/driver modifications and other similar points.

No comments: