Zoltan-

We also run TSM on a dedicated Linux/Intel, although on a smaller scale
than you.  The server has 2.5GB RAM and two 2GHz Xeons.  Our DB is 23G
of which 70% is in use. Buffer pool is set at 32M (8192 pages) with
selftune on. Expiration is run daily - typically completes in 4-7
minutes!  

I'm curious why your experience is vastly different.  Are your db
volumes part of a file system or raw volumes?  Ours are file system
volumes (ext3).  Given Linux's propensity for using unallocated RAM as
file system cache and the generous memory this server has, a
significantly higher number of DB pages may be in memory than meets the
eye.  Using raw volumes bypasses this caching. Perhaps the Linux file
system cache management is more efficient than the TSM buffer pool
management - maybe the adage 'less is more' in terms of TSM buffer pool
applies here? 

How many objects get processed during your expiration runs?  We're in
the magnitude of 10s of thousands of objects.

-Ken

-----Original Message-----
From: ADSM: Dist Stor Manager [mailto:ADSM-L AT VM.MARIST DOT EDU] On Behalf Of
Zoltan Forray/AC/VCU
Sent: Monday, April 28, 2008 10:41 AM
To: ADSM-L AT VM.MARIST DOT EDU
Subject: DB Bufferpool sizing - continued


Pertaining to the recent discussion of buffpool sizing (larger vs
smaller
- cpu usage, etc) I would like to get some opinions.

I followed the discussion and was aware of the issue of going too large
and killing any benefits by increased CPU usage and such.

So, I have been experimenting.

My big Linux TSM server has 8GB RAM and is dedicated to TSM, so other
products using/sharing the memory is not an issue.  The TSM DB is 160GB
assigned with 77% used.

The buffpoolsize (with selftune) was set to 768MB.

I just bumped it to 1GB (still at the 1/8 of real memory guideline) to
see if it would improve things, using EXPIRE INVENTORY as a benchmark
and trying to ignore the "Cache hit%" value (which I could never keep at
or above 99%).

Before the change, I was seeing EXPIRE INVENTORY running 46-51 hours.

This past weekend, it ran 40H.

I realize that having only 1-test result after the change is hardly a
good indicator of either positive or negative results.

So, my question is this.

Do I bump it to 1.25-1.5GB (still well below the "guidelines" of 1/8-1/2
of real memory) and see if that improves things even more or do I drop
down to 512MB to see if it improves things even more than the memory
increase?

Anyone else willing to share their DB sizes/buffpool/expire inventory
durations?

Open to all thoughts (besides the need for another TSM server......this
is already in the works and we have stopped adding new nodes to this
server).