ADSM-L

Re: To Collocate or Not to Collocate?

1998-01-05 10:13:26
Subject: Re: To Collocate or Not to Collocate?
From: Andrew Raibeck <storman AT US.IBM DOT COM>
Date: Mon, 5 Jan 1998 10:13:26 -0500
In large ADSM installations, tape storage pools can grow to hundreds or
thousands of individual volumes. Without collocation, it is conceivable that
data from a client could be spread across many of these tapes, mingled with
data belonging to other clients. This means that if you had to restore a large
amount of data (i.e. an entire client, or even a fairly large file system or
directory), the potential for a large number of tape mounts exists.

For instance, suppose the data is spread across 100 tapes and you have a manual
tape environment where human intervention is required to mount a tape. This
means 100 manual tape mounts. Even if it takes only two minutes to mount a tape
(and this is pretty optimistic in a manual environment), that's 200 minutes or
three and a third hours spent just mounting tapes! In an automated environment,
if tape mounts average 30 seconds (newer tape robotic technology may have
improved on this time), that's still 50 minutes spent just mounting tapes.

With collocation, ADSM attempts to keep each node's (client's) data on its own
set of tapes. However, even with collocation by node, the problem mentioned
above scales down to the file system level. That is, if you have a lot of large
file systems on a client, collocating by node may still result in a large
number of tape mounts if you had to restore the entire file system. As of
version 2.1.0.10 or 2.1.0.12 (I forget which), collocation by filespace was
introduced. With collocation by filespace, ADSM attempts to keep each client
filespace's data on its own set of tapes. Thus if CLIENTA has three file
systems, /usr, /home, and /development, then ADSM will try to keep all backup
versions for /usr on one set of tapes, all backup versions for /home on another
set of tapes, etc.

Collocation is set at the storage pool level via the DEFINE STGPOOL or UPDATE
STGPOOL command. Just specify one of these options: COLLOCATE=NO (disable
collocation), COLLOCATE=YES (collocate by node), or COLLOCATE=FILESPACE
(collocate by filespace).

The drawback to collocation (there is no free lunch :-)   ) is that on the
backup and storage pool maintenance side of things, you will almost certainly
incur more tape mounts and use more tapes. For example, suppose you have two
clients with two file systems each. Further, let's assume for simplicity's sake
that the combined data on these two clients would exactly fill a single tape.
Without collocation, you could mount one output tape during backup (or
migration) and all the data would be written to that tape. With collocation by
node, you would need two tapes (one per client), and each tape would be only
partially full. With collocation by filespace, you would need four tapes (one
per client filespace), and each tape would be even less full than with coll
ocation by node. Yes, as time goes by, the entire tape capacity could be used.
But the additional overhead still exists.

So to collocate or not is a decision you need to make based on what resources
you have at your disposal to maintain a collocate storage pool, and the
importance of speedy restores. Initially it sounds like too much overhead to
collocate when you probably do large restores only once in a great while
(hopefully!). However, when you *do* need to do a large restore, you'll
probably want it to complete yesterday! So that is something to take into
consideration when you decide whether to use collocation.

Alternatively you could set up, say, two tape storage pools: one that is
collocated and one that is not collocated. With ADSM's policy-driven
management, you can configure non-mission critical clients to go to the
non-collocated storage pool, and mission critical clients to go to the
collocated storage pool.

Hope this helps,

Andy Raibeck
ADSM Level 2 Support

        ADSM-L AT VM.MARIST DOT EDU
        01-05-98 09:00 AM
Please respond to ADSM-L AT VM.MARIST DOT EDU @ internet

To: ADSM-L AT VM.MARIST DOT EDU @ internet
cc:
Subject: Re: To Collocate or Not to Collocate?

> -----Original Message-----
> From: Mark Linehan [SMTP:markl AT NSMC.PARTNERS DOT ORG]
> Sent: Monday, January 05, 1998 8:19 AM
> To:   ADSM-L AT VM.MARIST DOT EDU
> Subject:      To Collocate or Not to Collocate?
>
> Big question, I finally got reclamation working on our system. It's
> amazing
> how many tapes change from full to filling when you change reclamation
> from
> 60 to 30! Well, now that this is behind me, the next question I would
> like
> to throw out is "How usefull is collocation?" Should I enable it? Will
> it
> save me that much time on retores? How much longer will it take to
> perform
> backups with collocation enabled?? I have to A S S U M E it is going
> to
> take a long time for collocation to first take place when it is
> enabled.
>
> Mark
<Prev in Thread] Current Thread [Next in Thread>