[rescue] tape drive with no SCSI

[Bill Bradford]

On Wed, Nov 07, 2001 at 07:19:08PM -0500, Patrick Giagnocavo wrote:
> The real problem with IDE is that the drivers need to be properly
> written - sounds like (according to what I pick up on the web) that
> hasn't happened for Solaris.

Nope.  Occasional data corruption when using UFS Logging + IDE drives
under Solaris 8; thats one of the main reasons the new SunHELP server
is all-SCSI.

I could type "sync" on the old machine and it would sometimes take
10-15 seconds to come back to a prompt.  *continually*.  I do that on
the new machine:

\u@\h:\w> time sync

real        2.3
user        0.0
sys         0.1
\u@\h:\w> time sync

real        0.0
user        0.0
sys         0.0
\u@\h:\w> time sync

real        0.0
user        0.0
sys         0.0

Etc.  Never more than five seconds.

Anyway, with UFS logging, If I did, say, a large tar -zxvf (like the GCC
source code) and then tried to IMMEDIATELY read/write a file that was in
the archive, sometimes I'd get corrupted data because the system (and drives)
couldnt keep up with playing back the UFS logs.  This was with a 300Mhz AXe-
based OEM board, 256K L2 cache, 128meg RAM, and dual Seagate/Sun Barracuda
ATA-100 IDE drives.

If I wanted to do anything with semi-large files, I'd have to untar/gzip them,
"sync", wait 10 seconds, "sync", "sync", "sync", etc, till I was sure that 
the logs had been played back - then I could get things done.  Never had this
problem with non-UFS-logging, though.

Lost a year or so of old mail due to this "problem".  IDE sucks for server
use, but is just fine and dandy and cheap in a single-user workstation where
you're not hitting the disks a lot.

Bill

-- 
Bill Bradford
mrbill at mrbill.net
Austin, TX