Re: Benchmarking scripts for 9i/10g

Jim Smith wrote:
> In message <1163780134.513321.71930_at_m73g2000cwd.googlegroups.com>,
> Charles Hooper <hooperc2000_at_yahoo.com> writes
> >Tarby777 wrote:
> >> Hi all,
> >>
> >> Are there any freebie scripts out there that can be used to generate
> >> benchmarks for an Oracle instance? Doesn't have to be too detailed,
> >> doesn't have to be perfect... just looking for something that does a
> >> fair chunk of work so that I can get an idea of the performance of our
> >> in-house testing rig relative to what our customers have...
> >>
> >> TIA
> >> Tarby
> >
> >* Statspack (ships with Oracle)
> >* AWR (additional cost feature for 10g Enterprise, turned on by default
> >even in the Standard Edition of 10g R2 - but can't be licensed for use
> >with that version)
> >* 10046 trace at level 8 or 12 (see "Optimizing Oracle Performance"
> >book by Cary Millsap)
> >* 10053 trace (see "Cost-Based Oracle Fundamental" by Jonathan Lewis)
> >* Directly query the performance views in Oracle
> >
>
> These are not benchmarking tools, they are performance
> monitoring/analysis tools. The OP was looking for something to generate
> a workload.
>
> I don't know of anything Oracle-specific but Mercury Loadrunner
> <http://www.mercury.com> gets a lot of mentions when these things come
> up.
> --
> Jim Smith

You are of course correct that the suggestions that I offered are not benchmarking tools, in the sense that they do not produce an artifical work load. I read the original post a bit different, quite possibly incorrectly.

If an actual work load is already present on the database instance, such as an ERP package that runs fast when in a development environment, but slowly when deployed to customer sites, it would make sense to use the ERP package as the load on the database instance, and use those tools that I mentioned to derive not only the performance characteristics of the two systems, but also to determine the motivating event that caused the witnessed performance characteristics.  If one were to benchmark a non-representative work load, would it be beneficial? Quite possibly, but also maybe not.

Charles Hooper
PC Support Specialist
K&M Machine-Fabricating, Inc. Received on Sat Nov 18 2006 - 06:36:16 CST