I have (seems to me) a very strange issue with Analysis Services

I developed a cube on an old, small server (2x PIII 866 with 768MB RAM running Windows 2000)

I copied this cube to a big production machine (2x Xeon 3.06 with 4GB RAM running Windows 2003 with /3GB tuning)

The cube is MOLAP

After pointing the queries to the new server, I notice that my queries take about 4x LONGER. I have tried re-running usage-based queries, and have gone through the registry settings trying to make them match between the two machines... I have also gone through the various performance tuning documents... but I can't seem to make the production machine any faster

The only thing I can think of is that the disk subsystem may be faster on the development machine. But while there's plenty of memory on the production machine, I can't seem to get analysis services to use much of it (it is only using about 300 MB)

I know this is pretty vague, but I wonder if there are "typical" things to look at when moving between machines, or when analysis services doesn't seem to be using enough memory during query time

Thanks

Jonatha

Registry Settings!? You should be seeing much more cache memory being used.
Looks like your new machine is using default values for memory usage. Look
under
Properties/Environment
Memory Settings
Min Allocated Memory
Mem Conservation threashold

See TechNet/MSDN for whitepapers on optimizing these settings.

hope this helps.

dlr

"Jonathan Levine" <myfoo (AT) spam (DOT) spam> wrote

gone through the registry settings trying to make them match between the two
machines... I have also gone through the various performance tuning
documents... but I can't seem to make the production machine any faster.
production machine, I can't seem to get analysis services to use much of it
(it is only using about 300 MB).
to be using enough memory during query time?

what is your disk subsystem and Raid config? (on the old and new server)


"Jonathan Levine" <myfoo (AT) spam (DOT) spam> a écrit dans le message de
news:FF8DA39B-F01C-457C-BE91-6A75895F852E (AT) microsoft (DOT) com...
gone through the registry settings trying to make them match between the two
machines... I have also gone through the various performance tuning
documents... but I can't seem to make the production machine any faster.
production machine, I can't seem to get analysis services to use much of it
(it is only using about 300 MB).
to be using enough memory during query time?

Dennis,

Thanks for your reply.

I have set these to 768 MB minimum, 2623 MB maximum. It's interesting to me that on startup, msmdsrv doesn't seem to get anywhere close to 768 MB... it's usually bumping along at 300-400 MB working set.

Jonathan

Jéjé,

Thanks for your reply.

"Jéjé" wrote:

The old server has the system drive and data drive on separate channels. The data drive is RAID-5 and system/swap are on RAID 1. The controller is an old Dell PERC-2. The disks are only 10K RPM, but the controller has a lot of cache.

The new server has a pair of 15K RPM drives is an IBM x335, using its on-board RAID controller configured as RAID 1. System and data are on the same channel. I'm not sure of the details of the x335's onboard controller, but I think it doesn't have any dedicated cache.

Jonathan

"Jonathan Levine" wrote:
One thing I've noticed is that the production cube server seems to be sending back lots of data to the client.
I have set Execution Location=3 and Default Isolation Mode= 1 on the client, but there doesn't seem to be
any difference in performance or amount of data sent back. I'm pretty sure that the locale id and SP version of
both server and client are the same, but I wonder if there's any way to trouble-shoot this (is there any log file
where the analysis server explains what it's doing and why?)

Thanks,

Jonathan