Earlier I wrote:
> I have a WangDAT Model 1300 tape drive (4 or 5 years old now), which
> is working fine on our IPC (running 4.1.3). I recently set up a
> SPARCstation-5 (with 4.1.3_U1) to replace the IPC as our main server,
> and put the tape on it. Unfortunately, the tape is no longer working
> correctly.
The problem appears to be that older SCSI devices cannot correctly
share the bus with a device that is running fast synchronous, as two
of our four disks can do (including the SS5's internal 535M disk).
Solving the problem apparently requires forcing the SCSI subsystem to
use asynchronous transfers only.
To do this under 4.1.x, run "adb -w /vmunix" and do "scsi_options?W 58",
quit and reboot. For a longer term change, I believe the right thing
to do is to remove the SCSI_OPTIONS_SYNC flag from the initialization
of scsi_options in scsi/conf/scsi_confdata.c. (Apparently under 5.x
you add "set scsi_options = 0x58" to /etc/system.) Having done that,
the tape appears to be working fine. Hooray!
Many thanks to Paul Kiesel for having forwarded a relevant
document to me from Sun; it is entitled "SCSI Configurations using
Single-Ended Fast SCSI Devices" and is dated October 21, 1992.
If anyone is interested, I can email copies of it (it's 20k long).
He also referred me to the comp.periphs.scsi FAQ (see, e.g.,
ftp://rtfm.mit.edu/pub/usenet/comp.periphs.scsi).
I'm not sure what the impact on performance will be. Testing with bonnie
on sd0 showed very similar performance (about 10% faster on block reads
with synchronous 10Mb/sec, but otherwise similar), but with multiple disks
active at once I suspect there may be some degradation of performance.
For the long term, I will either put in a second SCSI card for the slow
devices, or buy a newer tape drive that understands SCSI-2.
Several people suggested using active termination; we had, and still
have, a standard passive terminator. Since the tape drive is a Mac
peripheral, I have it chained in with a Mac-to-microSCSI connector
and a Mac terminator. While we have Sun-standard active terminators
around, we don't have Mac-standard ones; I'll look into getting one,
in case the passive termination is hurting performance.
Many thanks to the following for their suggestions:
Woobin Lee <woobin@fizzle.ee.washington.edu>
Paul Kiesel <kiesel@vishnu.ueci.com>
Ken Stewart <kstewart@ncr-mpd.FtCollins.NCR.COM>
Paul E. Begley <peb@netaxs.com>
Gary Field <garyf@wiis.wang.com>
netman@rave.com (Network Manager)
Jeff LaCoursiere <z056716@uprc.com>
Ross Margo <rambo@raw-55.Eng.Sun.COM>
Chris Metcalf, MIT Laboratory for Computer Science
metcalf@cag.lcs.mit.edu // +1 (617) 253-7766
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:09:09 CDT