Re: Slow SCSI Disk Access on AMI Elite 1600 controller

Ryan Shrout (linux-kernel@amdmb.com)
Tue, 14 Aug 2001 18:54:56 -0400


> > Timing buffered disk reads: 64 MB in 13.92 seconds = 4.60 MB/sec
> ...
> > 4 - 15K RPM Cheetah SCSI HDDs
>
> each disks will sustain 30-50 MB/s, so yeah, that's kind of low.
>
I changed the drives settings from Write-Through to Write-Back and it went
to 22.5 MB/sec

> > AMI Elite 1600 SCSI RAID controller with 64 MB SDRAM
>
> I wish people would start with relatively simple controllers,
> and only move to fancy ones if they know they need them,
> and can prove that the fancy controller works better.
>
> so for instance, these disks on any PCI ultra2 controller
> will easily sustain 100 MB/s. SW raid5 might slow things down
> if you had, say, a single celeron (260 MB's dram bandwidth),
> but certainly wouldn't effect a dual athlon with ddr.
>
I am using them in a RAID 0+1 configuration, so this kind of card was
needed. Besides, it was a gift (Free!) and I couldn't pass that up.

> > Red Hat Linux 7.2
>
> do you mean with RH's kernel!?!
>
No, actually, its 2.4.6-2smp custom compiled.

> > So, I then started trying to figure out to raise the buffered disk read
> > speed.
>
> do you have a simple, non-raid ultra2 controller around?
>
> > My only solution I came across -- find out if the SCSI
> > controllers/drives were in asynchronous mode and if they are, change
them to
> > synchronous mode.
>
> unless something dramatic is happening, the disks will stay in the
> same mode as they're detected at driver-init time, which should be
> right before your eyes in /var/log/dmesg. well, at least for a non-fancy
> controller. for a gold-plated controller like you have, you need
> to find some way to query the controller, perhaps a serial port,
> or some too provided by AMI. basically that's a non-linux question,
> since the controller has its own little OS that you need to interact with.
>
> > Now, how can I tell what mode my SCSI disks are in and how can I change
it
> > to synchronous if it isn't set that way already?
>
> unless your scsi cabling is utterly botched, they'll be synchronous.
>
Okay, that helps me so I stop looking for that setting! :)

Hopefully, going from 4.5 -> 22.5 MB/sec will be enough to temporarily stop
my mysqld crashing.

Thanks!

Ryan Shrout

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/