fullfeed related questions / cycbuff limitations

Russ Allbery rra at stanford.edu
Wed Sep 10 17:43:03 UTC 2003


Hendrik Scholz <hscholz at raisdorf.net> writes:

> Sep 10 09:37:13 foo innd: ME time 600077 hishave 2401(40789) hiswrite
> 126847(18236) hissync 0(2) idle 617(5332) artclean 847(18240) artwrite
> 209120(14910) artcncl 597(338) hishave/artcncl 14(338) hisgrep/artcncl
> 377(270) artlog/artcncl 17(338) hiswrite/artcncl 182(68) sitesend 0(1)
> overv 87907(14910) nntpread 64790(165159) artparse 89309(174979)
> artlog/artparse 0(0) artlog 1026(18264) datamove 4900(244713)

Your daily INN report will produce a much nicer summary of this, but it
looks like you're spending most of your time in disk I/O (artwrite
followed by hiswrite), which isn't surprising.

Tuning article writes is pretty much a standard disk I/O tuning problem:
try to distribute them across multiple spindles using the CNFS options,
have fast disks, investigate write-back caching in the drive controllers,
that sort of thing.

For history, having it on its own spindle may help.  Beyond that, INN is
writing out a random-access database, which makes it hard to get a lot
more speed out of it short of doing something like putting the history
database entirely on a RAM disk.

-- 
Russ Allbery (rra at stanford.edu)             <http://www.eyrie.org/~eagle/>

    Please send questions to the list rather than mailing me directly.
     <http://www.eyrie.org/~eagle/faqs/questions.html> explains why.


More information about the inn-workers mailing list