process causing large amount of disk transfer

[fthomas1973]

We have a SQL 2000 box - Compaq DL580, quad xeon, 4gb ram -connected to an Compaq MA8000 SAN (not fiber channel). We are implementing a third party data mining application. When we run a day's worth of data into the system our disk transfer/sec increases from spikes of less than 75 to a steady rate of 300 - 500. The number of transactions also increases. Although I have the data mining dbase on its own disk (data and logs on seperate disks) it causes other applications to be slow on their dbase activities.

Can anyone give me any insight on what to do to resolve this? Is there more information needed to take to the vendor and question their processes?
I am not an experienced DBA by any stretch so any input would be helpful. I have suspicions about the vendor's methods but not a lot of knowledge to be sure..

 

[Catadmin]

Sounds to me like either a processor issue or a disk controller issue. Have you verified the disk controller on your SAN is fairly new or that it can handle high I/O intensive tasks?

What about your SAN swap file? What size is it?

Run a Windows Trace while doing the input of data and see what your Processor is doing (and your RAM just to c.y.a.) when this happens to make sure it isn't either of them that is causing the problem.

Also, is SQL Server set to "Boost Windows Priority"? If so, it will take precendence over other applications running on the box.



Catadmin - MCDBA, MCSA
"If a person is Microsoft Certified, does that mean that Microsoft pays the bills for the funny white jackets that tie in the back???

 

[fthomas1973]

The processor on the server averages about 30-40% during the process. The sqlservr process has never used more than 500%(spike) of the total processor (800%). The controller on the SAN is a HSG80 with 256MB of cache and is supposed to be able to handle over 24K IOPS

SAN swap file.. I don't know what you mean by this.

I used the Idera sqlcheck app and perfmon to gather some stats. Memory pages never increased. I also noticed some lock timeouts occurring.

This is a dedicated SQL cluster. boost priority has been set.


Thanks for the ideas