And lo, Gadi Oxman saith unto me:
>
> On Tue, 13 Oct 1998 [EMAIL PROTECTED] wrote:
>
> > Hi all!
> >
> > I've been running raid-0 and linear modes on a disk array for some time. This
> > works really well! (I guess you allready knew :)
> >
> > Now, I moved the disks (six quantum fireball 6 GB scsi on two ncr53c810
>controllers,
> > three disks on each), into another machine and wanted to run RAID-5 on them.
I did the RAID-1 thing for a bit, then went to RAID-5 with three IBM
Ultrastar 2ES 2G disks.
> > The kernel which worked so well for me was 2.0.35 with md-tools 0.35.
> > The new machine runs 2.1.125, and I use the raid-5 capabilities in that kernel
> > unpatched. Furthermore, I use raidtools-0.41.
I use the latest(?) RedHat 5.1 kernel, with the contribbed raidtools.
> > I followed the instructions, and got a raid-5 array up running, or ``crawling''...
> > mke2fs takes long time (I went away from the machine, so I don't have timings).
RAID-5 with 3 disks was fast. But I upgraded to a 5-disk array (with two
more of the same disk, bought at the same auction no less) and now it
performs slowly. mke2fs spits out the initial numbers at the usual
speed, but then "counts up" rather slowly.
A clue: when under heavy I/O demand, the disks seem to wait, then dribble
out a little access, then access heavily for a time, then go back to
waiting. This started with the mkraid (and it's a 300 MHz K6, so CPU
shouldn't be the problem, and the second SCSI controller is a 53c8xx
just like the old one), which eked out a measly 5800 kbytes/sec because
it just wasn't keeping the disks busy somehow. It took ~30 minutes for
a ~8G array. mke2fs was also a lot slower on the array than on a ~7G
partition of my new 9G Viking II, on a K6-233/53c8xx machine as well.
> > What really pussles me is, that putc() seems to give good results, while block
>writes
> > are unbelievably slow. I'm aware that raid-5 will always be slower than raid-0, but
> > this is extreme.
>
> Ouch.. 100KB/sec on block writes is indeed unusable; strange, as we
> usually score very well on bonnie's block-writes.
I should see what bonnie says, but I hope watching the lights and
noting the poor mkraid performance are enough clues...
> Does it happen consistently on each run? Does one of the drives in the
> RAID array also contains an active swap partition? Does it also happen
> with a RAID set of 3, 4 or 5 rather than 6 disks on your system?
3 disks was great for me, but 5 sucks big rocks. Would I be better off
trying 4 disks tonight, or trying to cram the 0.90 RAID stuff into the
RedHat kernel, whose kernel-install process is not in the right place
(i.e. "make bzlilo")?
> > What is wrong ? Is raid-5 really so inherently slow, and has to lock the machine,
> > or am I missing something ? (The machine is a dual PII-350, I guess it should be
> > able to calculate the parity fast enough)
cat /dev/md0 > /dev/null &; top showed only 42% CPU utilization total, top
included. I wasn't watching the lights while I did this last, though.
Keith
--
"The avalanche has already started; |Linux: http://www.linuxhq.com |"Zooty,
it is too late for the pebbles to |KDE: http://www.kde.org | zoot
vote." Kosh, "Believers", Babylon 5 |Keith: [EMAIL PROTECTED] | zoot!"
www.midwinter.com/lurk/lurker.html |http://www.enteract.com/~kwrohrer | --Rebo
