Re: caching and benchmarking oracle?

I once did a performance test.
I did the following steps.

step 1
Try to identify the most common statements.

Step 2
Try to find out how many times a certain statement is executed. once a day, every day at 0900AM, during the whole day ??

Step 3
Use the info from step 2 and 3 to make some scripts that can simulate the behavior of an average user.

Step 4
Start make different versions of the "average user" script where you change the order of executing the statements.

Step 5
Start simulating different users. Start the different session, with different versions of the script.
Do not start all the scripts at the same time. Try to simulate the "real life login" behaviour of your users.

I ended up with 10 sets of SQL statements. Put some wait statements in your scripts. I started 10 session at the same time. ( each session with different version of the "average user" script ) Then I waited 2 minutes before starting another set of 10 session.

This gave me a good idea about the performance of the machine.

Hope this helps
Gert

In article <92c3vl$nsu$1_at_nnrp1.deja.com>,   gdas_at_my-deja.com wrote:
> Hello and happy holidays.
>
> Can anyone provide any advice on any relatively easy steps to take to
> benchmark the performance of an oracle database?
>
> My dilemma is that I need to procure hardware for a deployment but I
 am
> not entirely sure how much hardware I need. We have a development
> server right now used for a beta and it is pretty under-utilized. The
> problem is we expect the business to grow (hopefully) after the
> deployment so we expect our usage to increase. So I'm trying to find
> the high watermark for the current server and use that to help
> determine what to buy (and justify that purchase).
>
> This system is a read-only reporting system. So query performance is
> my main concern.
>
> We simulated several stress tests but are suspicious of them due to
 the
> fact that once things were cached in the SGA then simply repeating the
> requests over and over again didn't really provide realistic results.
>
> The problem is that different users login with different usernames and
> passwords and each user essentially gets a different constraint in all
> the where clauses of the queries.
>
> Right now I don't have an expensive stress test tool that is flexible
> enough to simulate different users doing different things at the same
> time. Can anyone recommend one?
>
> Or is there any way to temporarily disable the cache for this test?
> Or are there any formulas that anyone knows that I might be able to
 use
> to roughly extrapolate the load based on a limited scale test? The
> alternative is to basically run enough concurrent users (different
> users) until the machine is thrashed or the performance per session
> degrades significantly.
>
> This isn't my area of expertise, so I'm just trying to figure out how
> to do this. I appreciate any advice.
>
> Gavin
>
> Sent via Deja.com
> http://www.deja.com/
>

Sent via Deja.com
http://www.deja.com/ Received on Wed Dec 27 2000 - 02:40:27 CST