makehistory performance tests

Katsuhiro Kondou kondou at nec.co.jp
Wed Aug 9 06:53:40 UTC 2000


In article <200008082343.SAA07499 at darkside.norand.com>,
	Heath Kehoe <heath.kehoe at intermec.com> wrote;

} And here's the buffindexed results.  The -F option

This is very interesting.

} sorttype    options             # overview after one hour
} -------------------------------------------------------------------
} arrived     -O -x                    544451
} arrived     -O -x -l 10000           593206
} arrived     -F -O -x                 590870
} arrived     -F -O -x -l 10000        621499
} newsgroup   -O -x                    544451
} newsgroup   -O -x -l 10000           577456
} nosort      -O -x                    622958

Here is mine with three 2GB cycbuffs but they sit in
different spindles from a buffer for buffindexed.

sorttype    options             # overview writing performance
-------------------------------------------------------------------
arrived     -O -x -l 10000      58.4data/sec
nosort      -O -x               49.8data/sec
nosort      -O -x -F            62.8data/sec

# inndf -n
37037 overview data stored

'-O -x -F' invokes overchan.  This patch was posted
yesterday.

} >From these findings, I propose the default batch size
} be reduced from 100,000 to 10,000.  I also think
} Katsuhiro's OVNOSORT patch should be committed, since
} it yields the best results for buffindexed.

I'll commit mine, and default batch size also will
be changed like you said.
-- 
Katsuhiro Kondou



More information about the inn-workers mailing list