My original question was:
 >Is anybody out there running TWO (or more) dsmserv processes on the
same
 >machine?  If so, why (briefly)?
 >
 >I'm discussing whether my group should or should not do this,  and
would
 >love your input.
 >
 > I'll summarize the results to this forum: PLEASE RESPOND YES OR NO. to
 >lhmorris AT us.ibm DOT com.
 >
 >Thanks very much!

Thanks to everyone who responded.  I also called ADSM support and
discussed the issue with them. Here's a synopsis:

I got about 60-40 yes-no.  I imagine, however, that people NOT using
dual servers tended to ignore my question; ADSM support indicates that
most of their customers are using ONE server only (per machine, of course).

Under VM, ADSM does NOT multithread - it uses one CPU only.  So several
VM users had found it very advantageous to run multiple copies of
dsmserv, each on its own virtual machine.  ADSM documentation recommends
this - but only for VM.

Under AIX, ADSM DOES multithread all tasks, including expiration,
auditdb, and dbbackup.  ADSM support (level 2) knew of no performance
benefits to be gained by running multiple servers on the same machine.
Two dsmservs, each using 4 CPUS, should have about the same throughput
as one dsmserv using 8 CPUs.  The potential database I/O bottleneck is also
a non-issue, since the ADSM database volumes typically live on several
different disk drives.

Trevor Foley uses 2 dsmserv processes for security reasons.  One
dsmserv backs up nodes inside the firewall, the other backs up nodes
outside the firewall.  His concern was that an outsider might gain
access to the adsm server, guess the adsm password of a production node,
and thus be able to "restore" production data files.   I suggested that
his firewall could give him the same level of security without the need
to run two dsmservs.

Ben Bullock has one client node with 4 million files on it, and many
others of a smaller size.  Expiration spends a long time expiring this
one node, while  backups, restores, etc for the other nodes suffer.

He plans to run 2 dsmserv processes, one for the giant node, and one
for all the others.  Then he can schedule expiration only on weekends
for the giant node, but daily for all the rest.  This is a good reason
for running two dsmservs.  There's no way to say "expire inventory
node=xxx" - if there were, that would solve his problem.

Some other reasons I've heard for wanting to run multiple dsmservs are:

    "We want to split the network traffic: half the nodes on one
    netowrk card, half on the other."   But you can do this
    at the operating system level - ADSM doesn't care what network
    card the connection comes in on.

    "ADSM has performance problems when the database gets too
    large."  Certainly true for VM systems - used to be true for
    AIX but no longer (v 2.1.5.x).

    "We want to limit the downtime if we have to restore the ADSM
    database".  Well, that may be a valid reason if only ONE of your
    ADSM databases needs to be restored.  But if the whole machine
    needs to be restored, of course, you gain nothing.   We don't have to
    restore databases too often.

    For reference,  we backup an 8 Gig database in 2 hours on a J40
    (RS/6000 with 8 CPUs, 512 M memory). I gather it takes 1.5
    times as long to do a restore.

    "We want to speed up audit db, since we run it weekly."   If
    you have problems that REQUIRE you to run "dsmserv auditdb"
    weekly, you should resolve those problems.  We never run audit
    db unless we see certain errors, and that's only happened once
    to my knowledge, in six months of monitoring a dozen servers.

In my experience, the down side of dual dsmserv processes is:

    It's difficult to write scripts.  Our existing set of
    automation scripts query "the" server; if there were many
    servers, we'd have to re-vamp this automation.

    It's confusing to administer two dsmservs on the same machine,
    especially for someone who didn't set up the system.  This is not
    a petty gripe - the opportunities for human error are greatly increased.

    If the two dsmservs share a library, you have to be sure to
    partition the library correctly, or one dsmserv process may
    grab tapes which really belong to the other dsmserv process.

    You need two drives (to run reclamation) for every dsmserv
    process.  If you currently have less than four drives, you
    can't run two dsmservs and still do reclamation on both of
    them.

Again, thanks to all who responded.