I'm setting up a server that involves lots of database writing during regular updates, and I have wildly varying results between different machines. I'm trying to find out what I can expect from different machines (and hosting providers) without having to install the entire software stack to measure performance.

I've used hdparm -tT on the disks, but that measures sequential disk access.

Is there an equivalent test that's better for testing database-style random-access reads and writes? Or should I just rely on the manufacturer numbers?


Although SQLIO and IOMETER are great I strongly recommend iozone for detailed information in this area.

  • Bonnie is also a good test, but its always better to use multiple tools. Also remember the effect of cache on those tests. This is the URL for Bonnie since I fail to link stuff at comments: coker.com.au/bonnie++
    – coredump
    Jul 23 '10 at 16:45

You pretty much can rely on the factory numbers, but the problem is that you do not know what you talk about ;) Not in the negative sense.

what I can expect from different machines (and hosting providers)

Ok, simple: low performance as they save money. Point. I personally use a SuperMicro setup for a performance database with at the moment 8 discs, all 10k rpm. I doubt a hostingp provider will over that without serious pushing.

More complex: This will vary widely (your idea to measure is right), but even then you are not along on the system - so the performance can again ary a lot more depending what other people do at the time.

You have rightly identified the bottleneck being the discs.

SQLIO is my favourite. Remember to give the tool enough large a test file to actually move the head a lot.

  • Sorry, maybe "hosting providers" is the wrong phrase - I'm only looking at dedicated servers, so the machines aren't shared with anyone. Jul 25 '10 at 15:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.