Subject: Re: Seagate SCSI write-back drive cache - BEWARE!
To: None <netbsd-users@NetBSD.ORG>
From: Laine Stump <laine@MorningStar.Com>
List: netbsd-users
Date: 03/24/1997 14:50:07
> >> >./scsi -e -f /dev/rsd0d -m 8 -P 3
> >> 
> >> >and change WCE to "1".
> >
> >Just how safe is this even in the case where you have a UPS? Does the
> >drive properly send the sector to a spare if it encounters an error
> >while trying to write?
> 
> If you have AWRE turned on in mode page 1, yes.  Most drives ship with
> it turned on, but you should make sure.

I used the scsi utility to turn these two parameters on in both of my
striped ST15150W drives. The result was that file system writes went
from 6.8MB/sec to 7.05MB/sec, and raw disk read bandwidth (as reported
in the logfile created by lmbench) went from about 11.5MB/sec to around
6.5MB/sec.

Of course I wanted my old numbers back. So I used scsi to change the
numbers back to their original values - raw reads remained the same, but
filesystem writes went from 7.05MB/sec down to 5.5MB/sec. Getting
worried, I rebooted - no change. In desperation, I power cycled the
machine - STILL NO CHANGE!!

AIIIIIIIIGGGGGHHHHHHHH!!!!

So now what do I do? I have an identical machine running an identical
kernel which is still doing nearly 12MB/sec raw reads, but this one is
screwed. Is there something the scsi program does in the background that
may be causing this (maybe setting a default value it shouldn't set or
something? I want my old performance back!