Overview speeds and methods

Christian Balzer chibi at gol.com
Thu Aug 15 00:39:30 UTC 2002



Hello,

Just a a follow up, I tried to use tradindexed and from the
way it was going during makehistory it looked to be both faster
and more CPU intensive (not a biggy here). Alas I never got it
to finish a makehistory run with that method because:

a) It ran out of file descriptors, looks like it doesn't close the
temporary files correctly, lsof was full of stuff like this:
---
makehisto  4665 news  272u   REG       33,6    4021364       217 
/var/spool/news/incoming/tmp/histKRL0K9 (deleted)
makehisto  4665 news  273u   REG       33,6    4174950       218 
/var/spool/news/incoming/tmp/histqyP5ch (deleted)
makehisto  4665 news  274u   REG       33,6    4073264       219 
/var/spool/news/incoming/tmp/histCWBYcL (deleted)
makehisto  4665 news  275u   REG       33,6    4077812       220 
/var/spool/news/incoming/tmp/hist9btz61 (deleted)
---
And when increasing the batch size and number of descriptors I ran into

b) Gee, those files really are staying around it seems, because after 2 
hours of running makehistory and being pretty close to the end:
---
 news at nnrp:~$ makehistory -O -F -I -x -l 100000
/usr/bin/sort: /var/spool/news/incoming/tmp/hisTGIseXF: write error: No space 
left on device
makehistory: Can't sort OverTmp file (/usr/bin/sort exit 2), Success
---
On a 4GB partition, with the individual (visible) temp files (2 of them) 
being ~40MB each.

So I'm back to buffindexed, which at least seems to perform a lot better 
with the new, fast and dedicated RAID I placed it on.

This is INN 2.3.2 (Debian Woody package).

Dewa,

Chibi
-- 
Christian Balzer        Network Engineer        Engineering
chibi at gol.com   	Global OnLine Japan/Exodus Communications K.K.
http://www.gol.com/




More information about the inn-workers mailing list