Patrick,

standard stuff is to try to spread your database and log volumes across
SCSI controllers.  Also don't allocate copy2 volumes on the same drives
that copy1 volumes are on.  Consider splitting your database to get each
down below 15gb -- something of a threshold number for this product.

Your experience of NT servers backing up slower than UNIX and esp.
Exchange is also typical for TSM: each file is a database update.  More
smaller files take much longer to process than fewer larger files.
Again you can optimize the log and database performance by spreading
(equal size) partitions across multiple disk controllers.  Don't use any
one disk for more than one database or log volume.  If you aren't a DBA
just talk to one about how concerned they are about spindle count!

Regarding the bufferpool: let TSM adjust it itself.  The option is
SELFTUNEBUFpoolsize.  Just set it to 'yes'.

BTW, I was out at BP Amoco in Houston a year or so ago for a competitive
run off between Legato and TSM.  At that time I was with MCBA/Innovative
Business Systems.  Are you in Houston?

Regards,

Steffan Rhoads

"Hagan, Patrick L" wrote:
>
> I can not seem to tune my Solaris server correctly. I have a Sun E450,
> 4-300mhz cpu's, 4GB of memory, and 2 GiGE ports, one for Unix, and one for
> the NT world. As I loaded this machine with clients my db cache got really
> bad, as low as 88% during backups. As you can imagine everything runs slow,
> like expire inventory never finishes.
>
> My db is 50GB using 74% and should get smaller in the near future. It's on 3
> - 18gb disk drives, one dbv each, and they are TSM mirrored. My plan was to
> raise the buffpool a little at a time until the db cache gets to 98%. Right
> now it is at 131027, with 4GB of memory I believe this is way to small. I
> raised the buffpool to 196608 expecting to see a small performance increase,
> at least no worse. From what I saw my Unix, Unix db (archive), and NT
> exchange clients stayed about the same. The weird thing is that the "normal"
> NT servers backed up 3x slower than the night before. It seems like the NT
> clients with the most files backed up the slowest. I tested this twice, and
> got the same results.
>
> Does anybody have a clue what's going on?  Does TSM just suck on Solaris? Is
> there some kind of kernel mod I am missing?
>
> Another thing I can't figure out is what size to use for the tcpwindowsize
> on my TSM server. I am Backing up both Unix and NT so I am trying 1024. If
> someone has a similar system as I do, can you send me your dsmserv.opt file?
>
> Thank you for your time,
>
> Patrick Hagan
> BP Amoco - Houston