Re: What do we mean by performance benchmarking?

On Tue, 04 Jul 2006 20:20:06 -0700, Dibur wrote:

> What is meant by performance benchmarking, how do we estimate the upper
> limits for a db created on a specific server enviornment.

1. Performance benchmarking has many different meanings. And many of the meanings are simply not very meaningful.

In general, I think the term is used to mean a time to complete a specific action of set of actions.

Benchmarks - the time to complete a specific action, and the conditions for the benchmark (hardware, memory, tuning parameters, phase of the moon) are usually saved to permit comparisons at a later date or under different conditions.

Some organizations, such as the TPC (http://www.tpc.org) have created a 'standard set of actions' to be used. This has allowed companies to compete for bragging rights to being able to complete the actions in the shortest amount of time - using advanced tuning techniques, faster hardware, and any number of tricks.

Bottom line with benchmarks is that YOU need to determine a set of actions that are meaningful to you. When things change (better or worse doesn't matter, just change), you can then attempt to understand the reasons for the difference - often in the hope to make things [even] better.

Relying on anyone else's benchmark to make a determination of anything is generally useless (which may be the reason why management wants to see the benchmarks from others.)

2) What do you man by 'upper limits of the DB'?

Operating system specific limits, when provided at all, are provided in the appendix of the installation manual for the operating system. General limits are provided in the Oracle Database Reference manual.

These limits relate to size capabilities, not performance.

Performance limits may be guesstimated by looking at the TPC results. However, the TCP results are the result of very specialized tuning by professionals for very specific purposes under very specific conditions and for vvery specific durations. Deviation from any of the operation environment conditions, data set size, or expertise of the tuners will cause different results.

-- 
Hans Forbrich   (mailto: Fuzzy.GreyBeard_at_gmail.com)   
*** Feel free to correct me when I'm wrong!
*** Top posting [replies] guarantees I won't respond.