ADSM-L

Re: ADSM Restore Timings

1997-04-10 13:25:55
Subject: Re: ADSM Restore Timings
From: "Prather, Wanda" <PrathW1 AT CENTRAL.SSD.JHUAPL DOT EDU>
Date: Thu, 10 Apr 1997 13:25:55 -0400
My experience is on a configuration not exactly like yours, but since
you didn't get a lot of other responses here goes...

We run the ADSM server for AIX on an RS6000.  We are using STK 9710
silos with 3490-compatible drives.  (Comparable to the big STK silos for
MVS, just in a smaller box).  I have done a disaster recovery for a
Windows NT box  with about 6500 files, about 1 GB of space, where the
data was spread over 7 cartridges (not colocated), in about 40 minutes.

But I think the answer to your question is, "it depends".

I have seen individual tape mounts for ADSM take as long as 10 minutes.
The robot gets the tape mounted in 10 seconds, but it can take that long
to search all the way down the cartridge to the spot a particular file
is located.  10 minutes is about the worst I have seen.

Obviously, for important ADSM clients, using colocation is the best way
to go.  It causes things like migration and tape reclamation to take
longer and require more tape mounts, but these are back-end processes
that you will generally run outside your prime backup window, and if you
have robotic tapes you won't care.

I have no numbers for this, but I think the amount of time it takes to
process a recovery is not just dependent on the GB of data you have to
process, but on the number of objects (files, directories, ACL's) that
ADSM has to process in its data base.  There is going to be a data base
entry for every version of every object that is backed up.  I think the
amount of processing time required on the server is less dependent on
tape mounts than on the size of your data base and how much horsepower
you have available to traverse it.

In this list I have seen some people complain about the amount of time
required to process the ADSM data base for recoveries or reclaims, but
we have encountered no problems at all.  I can do a tape reclaim in 15
minutes.  But then again, my data base is still < 1 GB and ADSM is
running on a fast machine with very little contention from other
processes.  DO NOT SKIMP on horsepower for the ADSM server or DASD speed
for the ADSM data base.  Consider what resources your MVS machine is
going to have available during prime time when you are trying to do this
recovery - you would need to talk to someone with an MVS ADSM server to
find out how much ADSM requires in terms of CPU cycles.

Another dependency is the speed of the NT server you are recovering.
All the decompression is done there.  Restores on faster machines run
quicker than restores on slower machines.  In every case we have run
into so far, the client machine has actually been the limiting factor.

You also have the question of how fast you can pump 6GB across your
network.  It isn't just the speed of the token ring that matters.  Make
sure your network connectors (bridges, routers, etc.) don't present a
bottleneck.  On MVS, be sure to check out your gateway - it is not
unheard of for installations to have fast networks with relatively slow
gateways to MVS.

Hope this gives you something to work with.


 =======================================================================
=
Wanda Prather
Johns Hopkins Applied Physics Lab
301-953-6000 X8769
wanda_prather AT jhuapl DOT edu

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



>----------
>From:  William Kral[SMTP:wkral AT tne DOT com]
>Sent:  Tuesday, April 08, 1997 2:00 PM
>To:    ADSM-L AT VM.MARIST DOT EDU
>Subject:       ADSM Restore Timings
>
>We are looking at ADSM for our enterprise backup system.... since we are
>looking at this, not only for everyday type restores, but for entire
>directory and server disaster recovery, my main concern is with restore
>timings. From what I've read and heard, ADSM will do the job and is very
>reliable. With ADSM as a primarily incremental package, if we had an
>NTserver with 6 gig of data on a 16 meg token ring network being backed up
>to an ADSM MVS server, I'm concerned that after a few months it would take
>an extremely long time to recover the entire server.
>
>I would be interested on hearing anyone's experiences with major restores or
>directory restores using ADSM after the server has had a good amount of time
>to spread its incrementals over many tapes. We do plan on using tape silos.
>
>I have heard that a 60 meg directory that may take 10 minutes with a
>full/differential philosophy, which we currently use, could take a 6 or 7
>hour jump in restore time because the data is spread on multiple tapes.
>
>Is this fact or fiction is the question I have. I am aware of collocation...
>
>
>Anything would be helpful, thanks in advance.
>
>Bill Kral
>The New England
>WKral AT TNE DOT COM
>
>
<Prev in Thread] Current Thread [Next in Thread>