SUMMARY: Older SCSI Max. Record Lengths

From: Wes Pfarner (wrpfarn%tome@jezebel.ms.sandia.gov)
Date: Sat Nov 06 1993 - 02:07:33 CST


This appears to be a little-known area of interest as I only received four
replies. Thanks to the fab four for their time and kindness.

jpl@allegra.att.com (John P. Linderman)
kevin@uniq.com.au (Kevin Sheehan {Consulting Poster Child})
Martin Frost <me@CS.Stanford.EDU>
led@abend.cc.purdue.edu (Lew Doll)

--------------

Mr. Doll had a useful comment:

        The problem is the SCSI interface. It has a maximum block size of
126b in dd block sizes (126 512byte blocks). If you then plan to be able
to double buffer, then only half of that can be used so a SCSI transfer
can be going on while the next buffer is being DMA'ed from memory into the
buffer space in the SCSI controller. For example, format does not do double
buffering, so it does read/writes in 126 sector blocks.

--------------

Mr. Sheehan noted:

I would suspect the problem is either the driver choke point (minphys) or
the 16 bit limitation on some SCSI boards. The 4/200 does not have an
on-board SCSI chip as I recall, so you can replace the board. The 4/300
has a strange one (the sm).

I would look at your general upgrade path, or look at the possibility of
putting in one of the SBus based SCSI cards, in which the later versions
all have much larger counters. Cheaper than a CPU upgrade...

--------------

Other comments indicated that 8mm devices are perhaps smarter than I thought
as padding (IRGs - sort of) is minimized and is not likely to be a problem as
long as the streaming buffers are kept full enough. Also, at least two of the
replies indicated that they are writing records larger than 35 K with the same
kind of equipment.

Of course the major "other" comment is the age of the devices involved, we
are not likely to replace the CPU, but I am curious if anyone knows a source
for SBus attachments to the 4/280 VME bus - I haven't heard of such a beast.
Should anyone know this info, I would surely appreciate a pointer to the
source.

--------------

Original Inquiry:

We are experiencing difficulty in reading/writing 250 Kbyte records to/from
an Exabyte 8200 2.3 Gbyte tape drive. The host machine is a 4/280 with a
501-1217 Rev. 11 SCSI Host Adapter board. It is running 4.1.3 with no patches.

Testing has indicated that the maximum record length that is successful is
around 35 Kbytes. (34 x 1024 bytes works, 35 x 1024 doesn't).

We have also tried the same test on a 4/330 running 4.1.3 with no patches with
the same results, however it does work on a 4/490 running 4.1.2 with no patches.
We were able to do 250 Kbyte records with no problem.

We are using a third party SCSI device driver (Apunix's) on all machines, but
that seems irrelevant because it works fine on the 490.

The question is: How do we get 4200 and 4300 CPU boards to accomodate the
large record size? Is it possible to rev up the board, replace the board with
a 4400 CPU (expensive) or anything else?

The object in all this is to maximize the data space on an 8mm tape.

------------

Thanks for the help,

Wes Pfarner System Administrator Dept. 9236
Sandia National Laboratories Albuquerque, New Mexico 87185-5800
505-844-0684 wrpfarn@sandia.gov



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:08:28 CDT