Server benchmarking to compare performance

[brianoz]

The "Who makes the fastest servers?" thread made me wonder, does anyone else run benchmarks on their servers?

Reason I ask is that otherwise you don't actually know whether the fantastic exciting new server you've just spent a small fortune acquiring/leasing is actually faster or slower!

Had an interesting situation recently where two identical servers exhibited quite different performance characteristics. One was quite stable, the other tanked regularly when it was under any sort of pressure. My feeling was that it was an O/S difference - likely to be Centos 5.7. So in this case, identical hardware, but something introduced by a supposedly small O/S version difference made the difference between a viable and a problematic server.

I'm planning on keeping a library of benchmark results to allow me to keep track of servers from different providers and also to be able to identify servers that aren't performing.

Anyone else doing anything like this? Which benchmarks were useful? What do you do to keep track of server performance? Any interest in me sharing the benchmark code I have?

 

[concerto49]

check out - http://serverbear.com

It's a benchmarking site. Should help with what you are talking about.

 

[Collabora]

The "Who makes the fastest servers?" thread made me wonder, does anyone else run benchmarks on their servers?

I'm planning on keeping a library of benchmark results to allow me to keep track of servers from different providers and also to be able to identify servers that aren't performing.

This would be useless since you do not have access to the server and there are many other factors not controlled by provider.

 

[brianoz]

It's actually very useful, as I do have access to the server and I use the results to compare new servers to old servers. Or even, to compare a proposed server to a new server; the script will run and upload it's results to an area where I can see the results. That way I can give a single line command to a provider and ask them to run the benchmarks on an empty server (ie "wget -O - blahblah.com/benchmarkscript.sh | sh"), and then I can look at the results sent to me and form comparisons.

The benchmark needs to be run on an empty and quiescent server or the results have little value. That's in fact what I do.

You must have misunderstood me somehow? This is of course only useful where you have control of a server. If you're a reseller there's less value, though it would give you an idea of what a VPS can do.

The scripts gather basic config info too, that is, a 16Gb server will look better than a 4Gb server so you need to know that as well as other factors to compare meaningfully. Not at all an exact science, but the point is to have at least some idea.