Allen, absolutely agree with you that some nodes are worth collocating
and some aren't.

Collocation of your __offsite copy pool__ is what I'm talking about.
If you DID collocate your copy pool, with 100 nodes, you could get 100
tapes /day sent offsite when you did backup stgpool.

But if you DON'T collocate the copy pool....
and then you need to restore many nodes in parallel....
...you'll hit that tape-contention problem I described:
        tape 001 has data for nodes A,B,C,D,and E.
        node A mounts it for restore
        nodes B,C,D, and E  *want* to mount it but have to wait for node A to
complete.

Nick C says, about the tape-contention problem:
> Plan the restores out so they won't be
> stringing each other along.  If Node2 is dependent, operation-wise, on
> Node1, then plan to restore Node1 before you start Node2.

But I think I didn't make the above scenario clear to him.  It's not any
dependence on
business operations; it's that N nodes share one tape that's the
problem.

I don't have any good answers to this problem, and it comes up now and
then.
Ans 1: collocate copy pool. Yuck - sends too many tapes offsite.
Ans 2: collocate copy pool but just for business critical machines.
Maybe.
Ans 3: make portable backupsets often. Well, maybe. significant extra
work though.

It would be great if TSM would read the whole tape, and multithread it
out to the five clients simultaneously.  Has this been put into ADSM-R
as a recommendation?


"Allen S. Rout" wrote:
>
> => On Wed, 3 May 2000 12:26:59 -0400, Lindsay Morris <lmorris AT OPENMIC DOT 
> COM> said:
>
> > And another thing  :-}   ----
>
> > In the event of a site disaster, won't there be lots of tape contention?
> > [...]  (Unless you collocate your copy pool - not recommended.)
>
> Why is colocation not reccommended?  Certainly there are situations for which
> it's a pain, and pointless.   But in general?
>
> > Backup sets solves this one too.  But you'd have to make backup sets of
> > EVERYTHING on a regular basis.... Ouch!
>
> Think about this: maintaining colocation is functionally nearly equivalent to
> making backup sets of everything, all the time, except that you remove the
> intermediate step of mixing all the data together in the fist place.
>
> As always, plan this out:  How long can you afford to take to recover?
>
> At UF, Our campus authentication servers are worth the effort to colocate,
> ditto campus mail servers, web hosts, etc.
>
> Workstations of Data Center staff aren't worth it.
>
> - Allen S. Rout
> - NERDC TSM Admin

--
Mr. Lindsay Morris
Mr. Lindsay Morris
Certified: AIX,ADSM, TSM, HACMP,SP
Gresham Enterprise Storage
lmorris AT openmic DOT com
606-253-8000