Re: Discreet transactions?

Evgeniy G. Krylov (egk_at_e-krylov.spb.su) wrote:
: Folks! Can anyone clarify what is it - oracle's discreet
: transactions? And why there are no benchmarks with non-discreet
: transactions?
 

: Thanks!
: Krylov Evgeniy,
: STU, St.Peterburg, Russia

Hi!

Discrete transactions are difficult to explain. It is basically a mode for running transactions which greatly decreases the amount of work and code-path that Oracle has to execute. For example, discrete transaction does not do any logging. There are many restrictions associated with using discrete transactions (e.g. a single transaction cannot write to the same page twice) which makes it impractical for any real-life application that I can think of and I have personally not run into anyone that uses it.

However, the fact that discrete transactions has no practical use does not mean that it does not have any use. Oracle used the discrete transaction in order to attain exaggerated results in its heavily advertised TPC-A benchmarks. Even though the results are highly misleading - they do not reflect the true capabilities of the Oracle engine - Oracle continues to use them as a sales tool.

Discrete transactions are considered a "benchmark special" and because of this Oracle will be forced to retract the TPC-A benchmark figures in June. The Transaction Processing Council (TPC) was forced by public pressure to try to clean up its act after the shenanigans such as the discrete transaction became a public issue. Oracle can still try to show that the discrete transaction is not a benchmark special and I believe that they have until June to do this. In the meantime they are using the TPC-A to the hilt in order to sell their Oracle engine as the fastest one available.

The advice I give to my readership and seminar attendees is to ignore the TPC benchmarks and any discrete transaction benchmark. If possible do your own benchmarks. If this is not practical, use forums such as Internet or Compuserve to find out what transaction volumes other users are attaining. You will find that it is a magnitude or even two magnitudes less than the TPC benchmarks. These are more realistic figures to go by.

Hope this helps.

• Rich -- Richard Finkelstein Voice: 312-549-8325 Performance Computing, Inc. Fax: 312-549-4824 Chicago, IL Mail: rfinkel_at_infochi.com