On Sunday 05 April 2009 03:39:07 Frank Sweetser wrote: > James Harper wrote: > > Has anyone else noticed a performance regression in the last few weeks > > with the latest 2.5.x code? My backups are taking longer than they used > > to but I'm not sure what the cause is, it may well be completely not > > related to bacula... > > I didn't notice it before (all my tests run unattended), but I can see a > recent jump in the test results on regress.bacula.org. On April 30, the > 2.5.44 mysql test run on my test system took 57 minutes to complete: > > http://regress.bacula.org/index.php?project=bacula&date=2009-03-30 > > But on the 31st, it took 87 minutes: > > http://regress.bacula.org/index.php?project=bacula&date=2009-03-31 > > Does this look consistent with the jump in time you're seeing?
The timing of the regression tests is probably not a very good indicator, unless you use *exactly* the same regress files with different versions. This is because we often tweak regression scripts by adding additional output (not seen if it terminates normally), or by modifying the test (usually adding more tests or more testing), or as I recently did, I modified the CDash regression tests so that they *always* rebuild the full source (previously they only did so if the source changed), we have a good number of sleeps and a lot of waits in the scripts so the wall clock time can vary on the tests, and finally, the timing of the tests depends on the size of the Bacula source directory including all the objects and binaries (it is what is backed up), and the source is constantly growing ... All that said, we probably should design some good performance tests and run them on various versions of Bacula so that we are sure that nothing is going wrong over time. ------------------------------------------------------------------------------ _______________________________________________ Bacula-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/bacula-devel
