Some lucky individual (me), was able to get TSM servers with 32 GB of
memory.
All the documentaiton/redbooks from IBM/Tivoli on that I could find,
indicates that BUFPOOLSIZE should be tuned so DB cache hit rate is 98
and of course, common sense indicates higher hit rates are bettter.
There also
I saw something very like this. If I increase my bufpool size above
2GB, eventually TSM slows down to a crawl. I called IBM, and they tried
to send me to the performance group, so I just told them nevermind, I'll
put it back down to 2GB.
Matthew Glanville wrote:
Some lucky individual (me),
On Fri, 27 Jan 2006 10:11:55 -0500, Matthew Glanville [EMAIL PROTECTED]
said:
Unfortunately it appears that the TSM server, when using the database
buffer doesn't have very good method to find things in that buffer.
Thus searching through 30 GB of memory, or 10, or the class
Oo, neat!
Do I understand you correctly to say that you've got 32G of memory and
a 4G database? If you've got core of a similar size to your DB, then
I suggest an experiment: instead of sticking it all in a buffer, make
some RAMdisk, and stick a third copy of your DB vols there, and see if
I think the discussion could use more details about the architecture
being used.
If you haven't already, see IBM Technote 1208540.
Richard Sims
On Fri, 27 Jan 2006 12:05:54 -0500, Matthew Glanville [EMAIL PROTECTED]
said:
BUFPOOLSIZE when it is too large, which my guess is the point at
which a search through that much memory takes longer than a disk
read. The larger it it gets, the slower database reads get... This
mostly
I think the discussion could use more details about the architecture
being used.
If you haven't already, see IBM Technote 1208540.
Richard Sims
Ahh those details.
TSM server 5.3.2 on Solaris 9, 64 bit, 8 cpus' 32 GB ram.
DB size, 150 GB, 75% used
Current BUFPOOLSIZE 1 GB
My
On Jan 27, 2006, at 1:54 PM, Allen S. Rout wrote:
Sounds like that would make a fascinating chart.
And the recommended formula would have led you to something like 20G,
right?
With the really slow points, was the cache hit percentage good?
Something else that could be interesting is to set
