Our ADSM environment is as follows:

RS/6000 R30 4way - 512mb - 7134 disk pools & mirrored log
7137 RAID-5 DB, is 39gb in size, (70% utilized, cache hit is 97%)
3494 library with two 3590-B1 drives
AIX is at 4.1.4+  and ADSM is at 2.1.5.9+

Clients:

Sun and IBM.  Most are very large servers (over 200 gb). Client data
would include 10 million+ symbolic links, average file size about 40k.

We upgraded from an R20 with 2.1.0.7 -> R30 with 2.1.5.9,  5 weeks ago
and ever since this upgrade we have been unable to complete our backups
with in the window which previously worked.  This would not be a concern
except the time to complete the window has gone from 6 hours to unknown,
since we have NEVER completed a backup window. Instead we now have to
backup clients more selectively to ensure they complete.

It appears the problem lies when two or more clients begin "Expiring-->"
during incremental backups. This grinds every session to a trickle,
(maybe bursts of 40+ objects every 3-5 minutes, time shared between all
the active sessions).  During the time of slowdown, the server may write
less than 500kb every thirty seconds to the DB disks, and read less than
40kb in 30 seconds. This I/O rate indicates that this slow down is not
attributed to the DB residing on RAID-5 disk.  In addition, using "crash"
on the server we've NEVER seen more than one dsmserv process running
during the times of slow backup performance, yet when doing backups
when no expiring is happening we see multiple dsmserv's running (which
should be expected).  It appears there might be a serialization issue,
either with a server code segment or a db lock, however we can't seem
to identify specifically what the source is.

ADSM support and development are working on this problem, however I need
to know if anyone has experienced this situation, and if so what action
was taken.

Thanks!
Dan