To recap my issue - archives taking hours instead of minutes, randomly,
while other archives on same machine at same time finishing
appropriately...

Same problem as on AIX 5.1 from several years ago.  Gigabit ethernet
"smart card" mis-behaves when chksum_offload and large_send both turned
on.  I turned off both attributes, bounced the network interfaces, and
the problem is gone.

On AIX 5.1 the symptom was not slowness, it was file corruption, with
AIX 5.3 (almost 100% current - almost bleeding) the symptom was random
slowness.  Funny thing is that when the problem arose back on AIX 5.1,
the solution was 1) upgrade AIX driver code level or 2) turn off the
chksum_offload and large_send attributes.  I chose option 2) at that
time due to less downtime to the machine and application.

Seems if there was a driver fix for AIX 5.1, surely they would have
gotten it right by AIX 5.3...  :-(

-Tom

>>> Richard Sims <rbs AT BU DOT EDU> 11/10/2006 9:49:31 am >>>
Tom -

You'll likely have to do the standard analysis to isolate the
problem.  The best first approach is to watch the session from the
server, via repeated Query SEssion, and see what's happening during
it.  (You can get some sense of where waits were occurring in past
sessions by reviewing TSM accounting log data.)  If the problem is
constant on one or two files, perform a cp of the file to /dev/null
and see if that takes far longer than it should, which would point to
file system issues.  Whereas you say that other archives from that
client run at good speed, then it's probably not a network
configuration issue.  If still stumped, run a client trace.

    Richard Sims