Re: Oracle Profiler

On Tue, 20 Jul 2004 15:04:28 +0200, darkeagle wrote:

>
> Ok. Now I execute It. Results time for the procedure is 19.000 ms. I
> re-execute it. Result tims now is 2.000 ms...
> Briefly, on every execution the profiler time change in a considerbly
> manner...
> This make impossible to compare two execution of the same procedure with
> small changes at the procedure source code...
> Anyone has found myself results?

You machine does different things: handles network, runs batch jobs, handles email, prints and manages one or more sessions. Machine load is not constant, but changes over time. If you want to have statistically significant data, you have to perform several hundreds of runs, compute the mean time, the median execution time, the average execution time (this time "average" means arithmetic mean value) and the standard deviation. Data from one or two runs can differ wildly, but if you make several hundreds of them, they'll probably be grouped around one or two values. It also helps to check the time of day when testing. Is it a peek hour, when system is busy, and the guy next to you is trying to summarize 1TB of data for an annual financial statement, or is it the off-peak time. Testing a piece of software is a project. You have to decide what is the purpose of your testing, what methodology will you use, what resources do you want to devote to it and alike. Then you make a project plan, preferably in MS-Project or some spreadsheet and execute the plan to the letter, without cutting corners. Testing one or two runs, without a clear plan is like playing Wolfenstein, but consumes more resources.

-- 
A city is a large community where people are lonesome together.