On Sun, Jun 29, 2003 at 12:01:06AM +1200, Gareth Williams wrote:
> On Thursday 26 June 2003 17:16, Matthew Gregan wrote:
> > On Thu, Jun 26, 2003 at 05:00:42PM +1200, Carl Cerecke wrote:
> > > Is hdparm -t an accurate measure of hard drive speed?
> >
> > No. Not even close.
> Out of curiousity, can you justify this? Only, I sometimes use it to
> make comparisons between different drives, and it seems quite useful.
> Sure it's not a good indicator of filesystem performance, or anything
> like that that really matters, but for comparing raw transfer speed of
> two different pieces of hardware (ie. "hard drive speed", as per the
> origional question), how is this not a good measure?
Did you try and work it out for yourself first? He asked, I gave him
advice. If you really want to understand why it's a poor benchmark, you
should do some research instead of continuing to trust it in blind
faith.
Why hdparm(8) is not a good benchmark of "hard drive speed":
- it does not reflect real-world disk performance
- it does not run the test long enough
- it does not run more than one test
- it does not test both large and small I/Os
- it probably does not take caching into account
- it does not measure seek times
- it does not account for streaming versus interactive I/Os
- it does not reflect real-world disk performance (again)
There are other reasons, too, but I'm sick of listing them now.
Also note that the original question asked if it was an _accurate_
measure. It is absolutely not anywhere near an accurate measure.
> Secondly, while on the topic of hdparm... I notice I get
> significantly better performance with DMA mode enabled on my hard
> drive (not surprisingly!), which I can set with: hdparm -d1 /dev/hda
> (I had this in a script that runs at boot)
If the kernel has specific support for your IDE controller, it will
enable DMA correctly. You shouldn't be forcing DMA on with hdparm
unless you really know what you're doing.
> I recently upgraded my kernel from 2.4.10 to 2.4.21, compiled from
> source with basically the same .config as 2.4.10 had. It all works
> fine, except now when I try the above hdparm line (as root) I get:
> HDIO_SET_DMA failed: Operation not permitted
You should posted the entire output of dmesg when booting both kernels
to help us help you. My first guess is that the more recent kernel
lacks support for your IDE controller... and it turns out you've already
proven this is the case.
Cheers,
-mjg
--
Matthew Gregan |/
/| [EMAIL PROTECTED]
