Actually, it will work just fine. Any number of TSM server images on one
computer can share a physical tape drive connection (or tape drive
partition) with no additional contention problems, beyond the contention
you already have. They share quite nicely, giving drives to each other
as needed under the control of one of them that acts as library manager.

You're moving a LOT of data for that number of drives! I hope your
management understands that when you make this conversion, you may be
simply unable to move that quantity of data anymore. You may find that
you are no longer able to push data to the LTO and 3592 drives in
streaming mode, and that you suffer fatal shoeshining. This would cut
throughput to a tiny fraction of what it is now.

Are they going to replace their Cisco routers with Dell computers? I
didn't think so, and neither should they consider replacing their
specialty high-bandwidth AIX TSM systems. This system sounds like it's
already at the edge, hauling back the offsite tapes because you can't do
reclamation, and this will really slow it down to the point that it
won't be able to cope at all. You're going to find youself buying a
number of new tape libraries to spread the load out enough for a whole
herd of slow little Dells and/or dealing with a very complex SAN fabric.
Big, fast AIX (or Sun SPARC) systems would be much less expensive, even
in the short run. If money is tight, consider used IBM/Sun equipment.

P.S. Allen - congratulations on ufl.edu's big win on Monday!

Roger Deschner      University of Illinois at Chicago     rogerd AT uic DOT edu


On Thu, 11 Jan 2007, Zoltan Forray/AC/VCU wrote:

>
>The 4-LTO drives in one library, are shared amongst 3-TSM servers which
>transfer over 2TB, nightly.  The 4th TSM server is dedicated to Domino
>backups, which use it's 4-LTO and 4-3592 drives, 24x7, transferring almost
>3TB of mostly non-compressable data (I never get more than 800GB on a
>3592-2 tape) daily.
>
>With the drives being busy almost nonstop, this leaves little time for
>things like reclamation, etc.  Right now, I had to manually bring back
>150-offsite LTO tapes with less than 20% utilization, since I don't have
>time/drives to perform standard offsite reclamation by rebuilding them
>from the primary onsite tapes.
>
>It is a constant, hand-management juggling act, having to constantly
>intercede when the regular schedules get out-of-whack due to bursts in new
>data, LTO drives failing, etc.
>
>One of the servers is going to grow by 150GB or more of new data
>(radiological images), daily.
>
>So, trying to add one more TSM server to try to share 4-6 drives, just
>won't work.
>
>
>
>"Allen S. Rout" <asr AT UFL DOT EDU>
>Sent by: "ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>
>01/11/2007 10:04 AM
>Please respond to
>"ADSM: Dist Stor Manager" <ADSM-L AT VM.MARIST DOT EDU>
>
>
>To
>ADSM-L AT VM.MARIST DOT EDU
>cc
>
>Subject
>Re: [ADSM-L] Recommendations for hardware replacement/upgrade
>
>
>
>
>
>
>>> On Wed, 10 Jan 2007 12:08:22 -0500, Zoltan Forray/AC/VCU
><zforray AT VCU DOT EDU> said:
>
>
>> Thanks for the feedback.
>
>> Yes, I realize you can't beat AIX for I/O bandwidth.  Unfortunately,
>> it comes down to $$$$$$ (doesn't it, always).
>
>I think your $/performance-unit is much better on AIX than it will be
>in intel-land.  I call the x86 option cheap now, pricey later.  But
>you've already said that AIX isn't on the table.
>
>> I agree it would be beneficial to break things up. However, this
>> would lead to even more contention for resources (tape, tape
>> libraries) than we already have.  We have enough issues juggling
>> 4-TSM servers against 3-tape libraries (1-3494 2-3583).
>
>I don't understand how you concluded this.  Whatever the count of
>servers you're using, the drive use should be related to the client
>node count and behavior, and should not be varying too much.  Am I
>missing something?
>
>> I hadn't really thought about running multiple TSM server instances
>> on one machine.  Not sure if it is worth the effort/risk!
>
>If you are already running multiple TSM servers, you've got the
>coordination infrastructure in place already.  (or you don't in which
>case God Bless You) That won't be much different if you've got 2
>servers on one box.  I'm running 11 on one box now: Having relatively
>small databases makes a huge improvement in reliablity.
>
>
>- Allen S. Rout
>