db-derby-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Olav Sandstaa <Olav.Sands...@Sun.COM>
Subject Re: Performance regressions
Date Thu, 16 Feb 2006 15:37:28 GMT
Kathey Marsden <kmarsdenderby@sbcglobal.net> wrote:
> Right now we fly without a net for performance.  Does anyone run any
> regular performance regression tests, especially ones for which  they
> might be able to post results?

The results from the performance regression test I run are now
available on the following Web page:

  http://home.online.no/~olmsan/derby/perf/

This regression test runs "nightly" (Norwegian time) on the Derby
trunk. Currently only the following loads are included:

  * Single record operations: select, insert, update and delete of a single
    record

  * "TPC-B like load"

I hope to be able to including some join tests soon. I am also working
on trying to reduce some of "noise" from the test results (the
variance in the results are too large). 

The most positive change in the performance has been on Select
operation in client-server mode. Compared to the performance of the
10.1.1.0 release, Derby now performs almost 20 percent better on this
particular load (a CPU reduction per transaction of about 25
percent). This is mostly due to the performance improvements done by
Dyre, Knut Anders and Dan (and possibly others).

Please note that if you look at some of the runs from several months
ago there are data that indicates that there has been some large
performance regressions. These are mostly due to changes in either the
test clients or the test configurations (and should have been deleted
from the graphs).

Feel free to post any comments you have or proposals for improvements.

..olav

Mime
View raw message