> Was that 15% before or after updating from CVS?
> The more I think about the looping aspect the less I like it, so I'd
> prefer not to pursue making the unlock change for real.  But if it's
> really a 15% win then maybe we need to...
>                       regards, tom lane

After.  So far, I haven't been able to reproduce original the insert
problem.  A word of warning: the version I was testing with was patched
with some unapproved patches and that may have been part of the issue.

Here are my results (10k drive, NTFS):
fsync off, in xact: ~ 398 i/sec
fsync off, outside xact: ~ 369 i/sec
fsync on, in xact: ~ 374 i/sec
fsync on, outside xact: ~ 35 i/sec

with your code change:
fsync on, in xact: ~ 465 i/sec
fsync on, outside xact: ~ 42 i/sec

Don't put too much faith in these results.  If you are still
contemplating a code change, I'll set up a test unit for more accurate
results and post the code (my current tests are in COBOL).  


---------------------------(end of broadcast)---------------------------
TIP 9: the planner will ignore your desire to choose an index scan if your
      joining column's datatypes do not match

Reply via email to