January 21, 2005

More on ostress.exe

Continue from previous blog, I did not finished the performance
testing by using ostress.exe yet. As I read from command prompt
output, ostress.exe complained many failed logins. It seems even
though I directed the utility to login as sa with correct password, it
insisted to login as recorded users. And in read80trace.exe generated
.rml files there is no place for password, even though RML does
support node.

Moreover, to my surprise, read80trace.exe does not generate correct
DBIDs for most .rml files. Since we have unique login for different
databases, I pinned those wrong DBIDs at the first sight, I had to
modify them one by one. It is absolutely tedious while I had some 100
processes.

To ensure the same timing and waiting with production server, we have
to provide a control with delta replay mode enabled.

Sum it up, following have been done to make the performance comparism,
1. Changed passwords of all the users to the same as what of sa.
2. Set default password to the same password of sa in control file.
3. Enable delta replay mode in control file
4. Enter correct DBIDs for every .rml files. (select * from sysdatabases)

Here is a part of my sample.ini used as control file,
{Replay options]
Sequencing options=global sequence, delta, dtc replay
...
DefaultPassword=abcdef

What's more, according to my experience, don't use -O switch for
ostress.exe unless necessary. It will produce very large files, for my
case, it is 300MB for five minutes replay.

No comments: