Strange harddisk performance problem...

ichtys|dk · n00b Joined: 16 Sep 2003 Posts: 9 Location: Denmark

I've installed Gentoo Linux for my first time a couple of days ago, but one thing is not working proberly...

My system: Asus m2400n laptop, 1300Mhz Pentium M, 256mb DDR ram, 40GB harddisk...

My harddisk runs very slow, and it's pretty obvious that it has an effect on the whole system.

I've tried to recompile the kernel to ensure DMA support, but it doesn't change a bit.

I've run hdparm -X69 -c3 -m16 -d1 -u1 /dev/hda, and now hdparm -tT /dev/hda gives the following result:
/dev/hda:
Timing buffer-cache reads: 352 MB in 2.02 seconds = 174.26 MB/sec
Timing buffered disk reads: 10 MB in 3.42 seconds = 2.92 MB/sec

I've enabled DMA and everything... but still it works this slow...

What can I do, to improve the harddisk performance?

marshall_j · Posted: Thu Sep 18, 2003 1:47 am Post subject:

If you do the following what does your output look like?

ichtys|dk · n00b Joined: 16 Sep 2003 Posts: 9 Location: Denmark

I don't remember all the settings, but here is the things I remember for sure:

/dev/hda:
multcount = 16 (on)
IO_support = 32 bit (default 16-bit)
unmaskirq = 1 (on)
using_dma = 1 (on)
readonly = 0 (off)
readahead = 8 (on)

It should be okay, but it's still very slow....

Nermal · Apprentice Joined: 18 Apr 2002 Posts: 259 Location: UK

Make sure you have the correct IDE chipset support enabled in the kernel.

Tronic · Posted: Thu Sep 18, 2003 10:26 pm Post subject:

Mine are same otherwise, but readahead is 256! This is the default, I haven't changed it.

Might be the reason why I'm having very bad disk performance on random access. Running HP Xe4500 with 40 Go IBM Travelstar HDD.

hdparm "benchmark" results are 574 Mo/s and 19 Mo/s. I'm using the quotes because this is very bad benchmark for disk performance.

Tronic · Posted: Thu Sep 18, 2003 11:10 pm Post subject:

Hmm.. Changing the readahead setting didn't make much difference to bonnie++ results. I'm trying to find a low-level benchmark that works, so that we'll get some real results..

EDIT:

Looks like bonnie++ comes with somewhat useful separate program called zcav. With this, I'm getting 21 Mo/s at disk beginning. However, the results seem pretty unaccurate.