makehistory takes a long time on latest CVS
Richard Todd
rmtodd at servalan.servalan.com
Mon Aug 7 23:12:07 UTC 2000
In servalan.mailinglist.inn-workers you write:
>In article <200008070423.XAA06229 at darkside.norand.com>,
> Heath Kehoe <heath.kehoe at intermec.com> wrote;
>} Actually, I'm gonna run some tests, and find out which way
>} is fastest (i.e., shortest run-time of makehistory).
>As to tradindexed, overview needs to be sorted. It drops data
>whose article number is under lowest number - fudge(128).
No, tradindexed should automaticaly compensate for seeing lower numbered
articles, automatically triggering a rebuild of the newsgroup data/index
files as needed.
The reason for the sorting is that, unless you provide data to the tradindexed
system in something fairly close to sorted-by-newsgroup order, your performance
on stores is never going to be much higher than the aforementioned ~70/s, which
is enough to cope with the inbound traffic, but is not enough to make
a makehistory complete in a reasonable amount of time. It's strictly a
performance issue. The same seems to be true of buffindexed as well; I've
found in my experience that setting the sorttype to OV_NEWSGROUP in
buffindexed.c seems to make makehistory move along a good bit more
speedily.
More information about the inn-workers
mailing list