Fsys.aha8scsi w/29160 vs aha7scsi with 2940UW

QNX Newsgroups (Archives) qdn.public.qnx4
#1

We have two identically configured system in all respects except that one
was just upgraded to 4.25E and the Fsys.aha8scsi driver running with an
Adaptec 29160 while the other is running Fsys.aha7scsi with an Adaptec
2940UW controller. Otherwise all other factors between the machines are
identical including the disk drive model numbers, file system organization
and even the content of the drives (same data, executables, etc. etc., etc)

We upgraded to the 29160 with the expectation that this higher throughput of
the controller (40mb/s with the current drives) would offer better
performance in our databases. However, we found the exact opposite to be
true. The machine runninging the aha7scsi driver with the 2940UW dumps the
entire database in just under 9 minutes while the machines running aha8scsi
with the 29160 takes anywhere from 3 to 6 minutes longer to dump the exact
same database.

Needless to say we are very disappointed. Can you offer a suggestion as to
why the supposedly double disk throughput is producing an average of 50% the
throughput? Any suggestions as to what we might do to resolve the problem?
Disk I/O is becoming an significant issue.

K. Scott Piel
Senior Developer
Toptech Systems, Inc.

#2

Previously, K. Scott Piel wrote in qdn.public.qnx4:

Needless to say we are very disappointed. Can you offer a suggestion as to
why the supposedly double disk throughput is producing an average of 50% the
throughput? Any suggestions as to what we might do to resolve the problem?
Disk I/O is becoming an significant issue.

It sounds like there might be a problem with the new driver.
It is also possible that there is some kind of hardware
problem. At these high speeds you need to have SCSI
properly terminated with good quality cables. You also
don’t want to overtax the length limitations.

Even so I think that you should scale back your
expectations. The 2940UW controller runs at a maximum of
20Mhz X 16bits = 40MB/sec. The 29160, which I believe is
U2W, in theory could do double that. But now consider the
drive itself. If it is a really fast drive, it might put
out 12 or 14MB/sec sustained. The SCSI bus loading will
be lower, but the throughput probably not any better.


Mitchell Schoenbrun --------- maschoen@pobox.com

#3

Hi,

Can you provide the following info:


output sin ver
output from sin
output from sin arg
output from sin ir
output from show_pci -v
the make/model of controller
the make/model of devices
the layout of the SCSI bus (ie device/controller termination)
the settings from the SCSI BIOS
make sure Net isn’t using the tracelog ie Net -T,
increase the trace severity to 6 ie tracectrl -s6
post the output from traceinfo -R > log_file after
the database dump

Regards,

Joe
K. Scott Piel <spiel@toptech.com> wrote:

We have two identically configured system in all respects except that one
was just upgraded to 4.25E and the Fsys.aha8scsi driver running with an
Adaptec 29160 while the other is running Fsys.aha7scsi with an Adaptec
2940UW controller. Otherwise all other factors between the machines are
identical including the disk drive model numbers, file system organization
and even the content of the drives (same data, executables, etc. etc., etc)

We upgraded to the 29160 with the expectation that this higher throughput of
the controller (40mb/s with the current drives) would offer better
performance in our databases. However, we found the exact opposite to be
true. The machine runninging the aha7scsi driver with the 2940UW dumps the
entire database in just under 9 minutes while the machines running aha8scsi
with the 29160 takes anywhere from 3 to 6 minutes longer to dump the exact
same database.

Needless to say we are very disappointed. Can you offer a suggestion as to
why the supposedly double disk throughput is producing an average of 50% the
throughput? Any suggestions as to what we might do to resolve the problem?
Disk I/O is becoming an significant issue.

K. Scott Piel
Senior Developer
Toptech Systems, Inc.