Profiling nnrpd
Jeffrey M. Vinocur
jeff at litech.org
Tue Dec 17 15:42:56 UTC 2002
I ran XOVER in various batch sizes on a group with a fair number of holes,
with nnrpdcheckart: true, tradindexed, tradspool, repeatedly, and
calculated mean and standard deviation:
xover
range mean +/-
size
1 1587.1s 62.2s
10 158.9s 4.1s
100 16.2s 5.4s
1000 16.6s 3.7s
10000 23.5s 8.9s
38458 36.1s 19.2s
I suspect that the long times for retrieving the whole group at once may
have some confounding from the fact I used Net::NNTP which parses the
results of XOVER and returns them in a big hash, but I did attempt to
compensate by *not* letting the garbage collector dispose of the previous
hashes in the smaller cases.
I guess I should have included 20 in the set of sizes to try, but it's
reasonable to expect behavior similar to 10, I guess. So people with trn
should try rebuilding so it tries batches of 100 instead, perhaps. (If
anybody knows trn well enough to tell me how to do that, I can try it
myself.)
When I get a chance I'll try profiling the above and see where the time is
going.
--
Jeffrey M. Vinocur
jeff at litech.org
More information about the inn-workers
mailing list