Greetings,
A while ago I read a document on the 'net (which I can no longer find)
that claimed low values of conf-split (i.e. 1 or 2) were better on Linux
machines than the default 23. The rationale presented was that Linux does
a very good job of caching directory entries, and therefore conf-split
should be set at 1 for workstations and moderately busy servers, and
2 for very busy servers.
I've been trying to prove this -- I'm in the early stages of benchmarking this,
but queue injection at least seems to agree with this; I see either no
performance change or a small performance drop as conf-split increases from 1
to various values during queue injection.
I will write up the results, with graphs, when I am done. In addition, I
will release the (simple) tools I have written to do the benchmarking.
In the meantime, I would like input on the following: when testing queue
injection speed, how should I be limiting the number of parallel qmail-queue
processes? I've tried various maximums between 1 and 100, and see little
difference in the resulting queueing speed.
Should I just keep raising the maximum number of qmail-queue processes until
the machine begins to swap, then back off? With a maximum of 100 qmail-queue
processes on an old PC (P90, 64MB), I am CPU bound, not I/O or memory.
Charles
--
-----------------------------------------------------------------------
Charles Cazabon <[EMAIL PROTECTED]>
GPL'ed software available at: http://www.qcc.sk.ca/~charlesc/software/
My opinions are just that -- my opinions.
-----------------------------------------------------------------------
