We have had a similar set of circumstances here.  The failures are somewhat sp
oradic; at first we too believed it to be a problem in the network.  We had a
3172 which was running flat out -- the utilization on the box was well over
99%, but we were not using the offload feature.  (We are using MVS for our
server, and TCP/IP V3).  Mostly the failures occured during the regular day,
but a few occured during the night.  We scheduled the 3172s for upgrades, but
after the upgrade occured, we still had the problems.  We talked to the 3172
hardware folks, and they have been some help, but it still seems to be there,
although the frequency seems to be dropping somewhat.  We did increase the COM
MTIMEOUT value to 6 minutes also, and this seems to have been some help.

We had some other things we were looking at, too.  Our MVS dispatching priorit
y was set too low (below TSO and batch); we increased this, until it was just
below TCPIP.  At any rate, our MVS system runs at over 95% utilization on aver
age, and often much higher.   It is an Amdahl, and uses MDF and shares the
machine with other workloads.  We also were running with a very tight DASD
pool, causing clients to mount tapes for backups when they normally shouldn't.
 This also has been corrected over the past few days, although the pool
increases were not done in reaction to the TCP/IP problems.

For the moment, the problem seems to have disappeared, but we are not sure wha
t the "fix" was, if indeed there has been one.  We are still keeping an eye on
this, especially since others are seeming to have similar problems.

Jerry Lawson
ITT Hartford Insurance Group
jlawson AT itthartford DOT com