ADSM-L

Re: Collocation from nocollocation

2003-07-29 15:13:59
Subject: Re: Collocation from nocollocation
From: "Prather, Wanda" <Wanda.Prather AT JHUAPL DOT EDU>
To: ADSM-L AT VM.MARIST DOT EDU
Date: Tue, 29 Jul 2003 15:13:31 -0400
I agree, except that it's not necessarily one tape mount for each node.

If you have more clients than tape slots, you can still use collocation by
setting MAXSCRATCH on your storage pool to the number of tapes you want to
be used.

For example, if you have 200 clients, and set MAXSCRATCH to 100, TSM will
politely stack 2 clients per tape, so you would get 100 tape mounts during
migration, not 200.

If there are 2 clients on a tape, during a restore of clientA, it costs a
trival amount of time to skip over clientB's data (after all, you will be
skipping over some of clientA's inactive data anyway).  The point of
collocation is to save you TAPE MOUNTS  on restores.

By using MAXSCRATCH, you still get most of the benefits of collocation,
without blowing the capacity of your tape library or requiring a zillion
extra scratch tapes.

You just have to keep an eye on the number of tapes in the pool and adjust
MAXSCRATCH over time (as you add more data to your network) so that you
always have a sufficient number of tapes in FILLING status to benefit from
collocation.

I have never seen any "gotchas" from turning collocation on whenever you
want, as long as you keep an eye on MAXSCRATCH.  All collocation does really
is change the rules for the order in which tapes get filled:

If collocation is off, and TSM has data to write (migration or reclamation),
it looks for the least full FILLING tape.  If it doesn't have a FILLING
tape, it calls for a scratch tape, up to MAXSCRATCH.  If MAXSCRATCH has
already been reached, the process fails.

If collocation is on, and TSM has data to write (migration or reclamation),
it looks for the least full FILLING tape where there is ALREADY data stored
for the client.  If there is no FILLING tape with existing data for the
client, it calls for a scratch tape, unless MAXSCRATCH has already been
reached.  If MAXSCRATCH has already been reached, it takes the least full
FILLING tape available.  If MAXSCRATCH has already been reached and there is
no FILLING tape available, the process fails.

That's all there is to it.

And in an automated library, I have never seen any benefit to have MOUNT
RETENTION set higher than 0.

Wanda Prather
Johns Hopkins University Applied Physics Laboratory
443-778-8769

"Intelligence has much less practical application than you'd think" -
Dilbert/Scott Adams






-----Original Message-----
From: Rick Saylor [mailto:rsaylor AT AUSTINCC DOT EDU]
Sent: Tuesday, July 29, 2003 1:09 PM
To: ADSM-L AT VM.MARIST DOT EDU
Subject: Re: Collocation from nocollocation


Bill,

I made this change a while back without any real problems. I just enabled
collocation. It took some time but eventually my tapepool became
collocated. Also, TSM shouldn't crash because it ran out of scratch tapes.
Remember, if TSM runs out of scratch tapes it will find the least used tape
and write on it. You loose collocation on that tape but TSM continues on. I
can think of only two situations that running out of scratch tapes would be
a major problem, 1-when a database backup is needed and 2-every tape in the
storage pool has become full. With a little planning and monitoring you
should be able to detect and take corrective action before TSM dies.

Also, if your drive mount retention is set to high you'll have a bunch of
idle tape drives. Remember, collocation will force a mount of at least one
tape for each node. So, make sure that you don't use the default of 60
minutes. Cut it back to say 5 minutes so that the drives will be freed up
quickly.

Rick Saylor
Austin Community College

At 11:22 AM 7/29/2003 -0400, you wrote
>Hi TSM'ers
>
>       Has anybody out there gone from 3 years of nocollocation and then
>changed to collocation for better restores.  We are currently at 2-3+ TB of
>info on 540 tapes (this is all in our 1 TAPEPOOL).  We have about 410
>scratch tapes and the uppers are worried about making the global switch to
>collocate will cause TSM to crash due to no scratch tapes.  Also if there
>are any other gotcha's I would be greatly appreciated.  Maybe even send
>you, some of our greatest pizza haha.
>
>Thank You,
>Bill Rosette
>Data Center/IS/Papa Johns International
>WWJD


--------------------------------------------------------------------------
Rick Saylor                Austin Community College   Voice: (512)223-1182
Senior Systems Programmer  9101 Tuscany Way           Fax:   (512)223-1211
Information Services       Austin, Texas  78754