Verify server performance

Question

I'm looking for a quick and SIMPLE way to verify that new servers are performing as expected. The most important metric is disk performance, second is network performance. I’m trying to prevent problems caused by misconfiguration of RAID arrays, NIC teaming etc. The solution should work with both physical and virtual servers. I don’t need sophisticated analysis with different workloads, just one set of benchmarks which I would run against a reference server and later compare to new ones.
One problem is that most benchmarks are not giving accurate results when running on a VM.

Answer 1

You haven't mentioned what your OS is on the server, so the solution may vary based on the OS. I'll pick some tools for windows-based systems since that's what I'm familiar with.

Try a variety of disk benchmark tools such as: HD Tune, [ATTO Benchmark][2], and [IOMeter][3].

For networking, I've only played with one specific tool called [Netperf][4]. Other's may have better suggestions, but I've found this tool to be helpful in giving me some basic and detailed information on my network performance for a given system. I've also done some manual and scripted file transfer tests timed tests with different data types to give me an idea of the network performance. I've done this with RAM drives on both sides of the client/server to remove the aspect of disk I/O as the limitation (to a certain degree).

The key thing I would like to point out is that you may want to spend the time and investigate workload patterns so that you can fine tune your disk subsystem to work the best for your environment. I think IOMeter can help you a lot with this once you spend time understanding the workload. The problem is that there is no one single answer or performance metric to answer this specifically for you (or anyone). There are some generalities with having high performing drives (low latency) and the correct RAID level for a given task.

The simplest thing you can run is a series of tests with these benchmarks to compare to your reference server (as you mentioned). The results don't have to mean anything specific to you, but as long as they're within a few percent of your reference, it's a nice way to ensure there is nothing drastically wrong/different in the configuration.

ps: I tried to link to all the tools I mentioned, but this system won't allow for more than one URL linked for spam reasons.