ADSM-L

FW: IBM3590 SCSI PERFORMANCE

1996-01-16 09:22:00
Subject: FW: IBM3590 SCSI PERFORMANCE
From: "PITTSON, TIMOTHY" <PITTSON1 AT BWMAIL1.HCC DOT COM>
Date: Tue, 16 Jan 1996 09:22:00 EST
This discussion about 3590 performance, or how fast ADSM can drive them, has
raised some concerns for us.  We're getting ready to make a proposal for
moving ADSM off of MVS and onto an RS/6000.  We narrowed down our tape
choices to 2 robotic solutions -  a DLT library (ATL ACL4/52 with 4 DLT4000
drives) or an IBM 3494-L12 with 2 3590-B1A tape drives.  I chose the IBM
solution because of the outstanding performance characteristics (and
reliability, I hope)  of the 3590 tape drives.  However, if ADSM can't
'drive' the 3590's any faster than the numbers quoted here, I'm wondering if
I'd be better off with the ability to run more concurrent processes with 4
DLT drives as opposed to the much faster (and fewer) 3590 drives.

A question for IBM - is ADSM performance with 3590 tape drives something
that's going to be addressed ??

Tim Pittson
pittson1 AT bwmail1.hcc DOT com

 ----------
From: owner-adsm-l
To: Multiple recipients of list ADSM-L
Subject: Re: IBM3590 SCSI PERFORMANCE
Date: Wednesday, January 10, 1996 1:24PM

Matthias,

We have 3590 SCSI connected drives to an R24. The best 3590 transfer rates
we have observed have been with BACKUP STG (disk to tape, or tape to tape)
achieving 4.5 MB/sec. More typically we see rates 2.0MB/sec - 3.0MB/sec.
Migration from disk to tape is usually slower.

The profile of filesize is very important in determining performance. Here
are
just a few samples of real situations that we have observed, moving about 1
GB
of data each time:

        Average filesize        BACKUP STG      MIGRATION
        =================================================

                 60 KB          2.4 MB/sec      1.4 Mb/sec
                100 KB          3.4 MB/sec      2.1 MB/sec
                150 KB          3.6 MB/sec      2.6 MB/sec
                850 KB          4.5 MB/sec      3.9 MB/sec

There are several ADSM parameters which greatly affect the performance of
data throughput. These figures were taken when the ADSM DB is on 7137 disk
array; ADSM LOG on SSA; storage pools on SSA. LOG mode is ROLLFORWARD.

Also, DB buffer pool size = 6144; LOG pool size = 1024; DB size was 2 GB;
TXNGROUPMAX = 256;  MOVEBATCHSIZE = 1000; MOVESIZETHRESH = 500.

File sizes less than 60KB are bad news; I have some figures taken before the
parameters were set as decribed - but more like 1 MB/sec.

It is my understanding that 3590 performance could be improved if a much
larger buffer size were to be used by ADSM. I believe 32KB is the current
value used and that at 64 KB rates of 7 MB/sec could be achieved; at 128 KB,
9 MB/sec - assuming other things could keep up ...

Another question, for IBM, could anyone confirm that ADSM (2.1.0.3) is
making good use of 3590 addressing capability to optimise file restores
and retrieves. What I mean is, is the 3590 track taken into account when
determining the restore/retrieve order for a set of backup/archive files.

Thanks.

Regards, Sheelagh
 ----------------------------------------------------------------------------
 --
Sheelagh Treweek                         Email:
sheelagh.treweek AT oucs.ox.ac DOT uk
Oxford University Computing Services     Tel:   +44 (0)1865 273205
13 Banbury Road, Oxford, OX2 6NN, Uk     Fax:   +44 (0)1865 273275
 ----------------------------------------------------------------------------
 --
<Prev in Thread] Current Thread [Next in Thread>