HDD benchmarking

Spida · Posted: Mon Oct 06, 2003 6:36 pm Post subject: HDD benchmarking

I heard that hdd-benchmarking results of hdparm were somewhat unreliable.
Since bonnie++ is can be configured to test really big reads/writes on the hdd, I have been using that for a while now. That way, it is possible to see what performace is seen by the application (and the user) and and not what performance the hdd may theoretically be able to deliver.
Has anybody experiences with this? Different results from bonnie++ and hdparm?

meowsqueak · Posted: Mon Oct 06, 2003 8:14 pm Post subject:

It makes sense that they should differ - as you suggest, hdparm simply tests the interface throughput at 'saturation' and doesn't make any allowances at all for real-world usage. You could probably consider the hdparm result as an upper bound on throughput.

I've run bonnie and hdparm benchmarks on a few of my drives. Bonnie, IIRC, tests the filesystem rather than the disk itself, so it's more useful for comparing, say, XFS with reiserfs. I did discover that XFS is really fast compared to ext3, but much slower when unlinking files.

So the two tools, in my opinion, test different things and they are both useful. Neither replaces the other.

taskara · Posted: Sat Nov 29, 2003 10:05 pm Post subject:

I want to use bonnie++ to test some external hdd's, can someone tell me how to use the command with an example? cheers!

meowsqueak · Posted: Sun Nov 30, 2003 12:28 am Post subject:

I think the default options work fine. Just run it on an existing filesystem.

taskara · Posted: Sun Nov 30, 2003 1:34 am Post subject:

I think you actually need to run it with options.

like

taskara · Posted: Sun Nov 30, 2003 1:41 am Post subject:

ahh ok..

just had to specify

taskara · Posted: Sun Nov 30, 2003 3:40 am Post subject:

here are the results:

USB

meowsqueak · Posted: Sun Nov 30, 2003 10:34 am Post subject:

At first glance, it seems the firewire is faster, but uses more CPU. I find the results a bit hard to read in my proportional mozilla font - if I had the time, I'd copy them to emacs and make a proper comparison, but that will have to wait.

Thanks for the info.

taskara · Posted: Sun Nov 30, 2003 1:29 pm Post subject:

nps.. yeah I drew the same conclusion.. which I found strange.. anyway.

if firewire IS faster, then it is copying more data at once, which stands to reason it would use more cpu.

u could average it out to find average usage vs time and compare that to usb..