ADSM-L

Re: TSM server performance help

2001-07-10 20:45:38
Subject: Re: TSM server performance help
From: Steve Harris <STEVE_HARRIS AT HEALTH.QLD.GOV DOT AU>
Date: Wed, 11 Jul 2001 10:46:12 +1000
It looks to me like hdisk4 is saturated with IO.  What's on it? DB? Logs? Disk 
storage pool? Sequential storage? One big conglomeration of everything?

You don't say what the storage server is. Maybe it could use more cache.
One of the problems with SANs is that AIX still sees them as simple disk 
devices, and therefore the serializations that apply to real disks are still 
used.
If hdisk4 is really big, it may be useful to redefine it as several smaller 
disks and use standard AIX facilities like striping to increase the number of 
concurrent IOs.
You could also separate your database/log/disk storagepools into several TSM 
volumes on different disks.

These days, a lot of shops make everything RAID5 without thinking about it.  
I'd debate whether RAID5 is useful for TSM diskpools.  I'm an old mainframer 
with DB2 experience, and there the trick was always to access your disks 
sequentially where possible.  The physics hasn't changed and RAID5 scatters 
your stuff all over the disks turning sequential access into random access.  
AIX will do smart read aheads were it detects sequential access and surely a 
migration is sequential in nature.  Can anyone with knowledge comment on TSM 
database access patterns? If these are often  sequential event the DB might be 
better off on JBOD disk. 

OK. so once you've got your data placement organized, run your test again.
If its still unacceptable, check the archives for tips about database buffering.

Regards

Steve Harris
AIX and ADSM (TSM on Thursday!) Admin
Queensland Health, Brisbane Australia

>>> Chuck Lam <chuck_lam AT YAHOO DOT COM> 11/07/2001 9:46:30 >>>
Server:  M80 running AIX 4.3.3, 4GB RAM, 768MB ps
         DiskStoragePool, Rlog, & Database on all on a
          SAN via fibre channel connection.
TSM version:    4.1

This is a newly configured system.  We are conducting
tests on it. When we tried to do a few backups by
themselves, they seemed to get done pretty fast.
However, when we had 4 migration processes and a
couple of backups going at the same time, everything
slowed down to a halt.  According to my client, his
backup transfer speed was about 100MB per minute. Can
anyone give me some suggestions to improve the
performance?  I have attached the 'topas' reading of
this server.

TIA


__________________________________________________
Do You Yahoo!?
Get personalized email addresses from Yahoo! Mail
http://personal.mail.yahoo.com/
<Prev in Thread] Current Thread [Next in Thread>