Hello all,

If you do not like to read long posts please skip this one - it can be
boring.

I have a very nasty problem at customer site and asking for your help are
we the only having it. If not can we fight it together.
Environment:
Server: v4.2.1.15, Windows 2000 SP2
Client: v4.2.2.1 (was 4.2.1.25), AIX 4.3.3ML09 + patches
Storage Agent: v4.2.1.15, TCP/IP or Shared Memory
SAN: IBM 2108-F16 (Brocade) switch + IBM 3583-L18 (ADIC) LTO with 2 drives
Storage policy: Stgpools use defined volumes and no scratches.
Yes, I know it's very unusual combination (Windows TSM server + AIX
nodes). Even support in the beginning thought it is AIX server w/ Win
client.

The node in question is somewhat mixed - small (<50kB) and large (Nx100MB
or 1-4GB) files. So for performance small files are directed to go over
LAN. Large files are bound to direct-to-SAN-tape class and storage agent
should take care of them. DIRMC is pointing diskpool as it ought to.

Now the problem:
1. Client starts backup, sent several directories over LAN, on first file
reported ANS1114I and waits server to mount a volume on behalf of storage
agent (OK)
2. Variable number of files are backed up for less than 1% of the volume
mounted (OK)
3. Client reports another ANS1114I (here is the problem).
Under the covers it seems than client requested this and next several
files to go over LAN instead of LAN-free. So mount is performed by server
for its own use (normal MgSysLAN path).
4. Another bunch of files are backed to second volume. 'Q SES' shows tens
of MBs so files go over LAN.
5. Third ANS1114I is reported and client resumes normal operation
6. Backup completes with no error, everything is backed up but for
enourmous time - test with 3-4 GB went good 20 minutes instead of 5.
The result of this SAN-LAN-SAN sequence is two volumes less than 5%
utilized, 5-minutes backup taking between 20 and 40 minutes and unwanted
data spread. Contents are
VOL1: file set 1, file set 3
VOL2: file set 2
And if accidentally second drive is busy (migration most often) the
process might go to bed in step 3 and wake up one hour later (when drive
becomes available on the end).

Partial workaround:
If DIRMC is set to point the tapes LAN-free works as it ought to and 4.3
GB take 5 minutes including mount time. But backup of 20kB in 5 files is
taking again 5 minutes just because DIRMC required tape mount. This moves
the problem in other area without solving it at all.

Support requested client upgrade from v4.2.1.25 to v4.2.2 with no success.
Upgrade of the storage agent (plus the server and other agents) is
postponed because we want to ensure this problem is resolved before to
introduce new ones in v4.2.2.

As always any hints, advices or similar experience are thankfully welcome.


Zlatko Krastev
IT Consultant