My situation:
PII-500/512MB
2 9GB scsi disks (one for /var/qmai/queue, one for the rest of the system)
user homedirs mounted over NFS (but that's not relevant here)
running stock redhat 1000fd 2.0.36 linux kernel (from RPM)
I'm running a small script that injects 1000 distinct messages, all destined
for either devnull@localhost or devnull@remotehost, in the localhost case
corresponding to a .qmail file with just '#' in it.
On this particular machine, concurrency never goes higher than 4 or 5 as long
as the script is periodically injecting messages. Looking at qmail-qstat
periodically during the run shows the number of unpreprocessed messages (aka
in the todo/ queue) growing to about 130 at the moment the 1000th message is
injected. Only then (when the script stops injecting) qmail-send suddenly
gains speed and concurrency does hit the roof, with the second number in
qmail-qstat quickly dropping.
Tests on other machines (our heavily loaded shellserver, or my homebox which
is not half as powerful) show very different results: qmail is not even
slightly impressed with the numerous injections and handles everything
gracefully, with a todo/ queue never bigger than 1 or 2 messages.
The bad results were observed on a heavily patched qmail, my first suspect
was the big-todo patch. Removing it made no difference. I tried a completely
vanilla qmail, with no difference either.
Anobody got any ideas? The machine is not swapping heavily.
Greetz, Peter.
--
Peter van Dijk - student/sysadmin/ircoper/womanizer/pretending coder
|
| 'C makes it easy to shoot yourself in the foot;
| C++ makes it harder, but when you do it blows your whole leg off.'
| Bjarne Stroustrup, Inventor of C++
