Question about ostress.exe

microsoft.public.sqlserver.tools

19 Mar 2007 5:05 PM

John T

Hello,

I have collected data with PSSDiag, and run the traces through read80trace.
I have a few questions regarding the use of the RML files generated from
read80trace when running the ostress.exe utility to simulate a workload. I
am running SQL2K sp4 on a WIN2K sp1 server.

1) Does ostress use the parameterized generic queries generated by
read80trace, or the actual queries caputured during collection, when you
specify '-m stress'? I ask this because I plan run a simulated mid-level
OLTP rate against a copy of our production system.
2) If the actual queries from collection are replayed, as opposed to the
generic queries generated by read80trace, how do I work around PK constraint
violations from the inserts?
3) The ostress test, in part, will be used to figure out IO capacity of a
new disk subsystem. If updates, deletes, and inserts no longer exist on my
restore of the production system, when I plan to run the stress test, how
will I guarantee a workload that simulates the actual physical IO associated
with the operations on production?
4) The production system runs transactional replication with queued
updating. Do I need to set up replication in our lab environment, if all I'm
looking at is the potential max IO capacity of one database?
5) How can I configure my xml config file to only capture activity for one
database? The documentation surrounding pssdiag capture indicates no way to
filter for a single database, and the options surrounding ostress show no way
to apply only the spids associated with a single database. Should I bypass
pssdiag.exe and just use a configurable server side trace?

Currently, we do not sync our distribution db backups with our published db
backups, so coordinating a stress test replay with replication seems
problematic.

Thanks to all responders.

John T