[Mailman-Developers] Performance under load (bursty message flow)

[J C Lawrence]

In running a number of lists at VA under mailman, I've noticed an
apparent performance problem: multiple simultaneous deliveries to
the same list performs badly.  Locking?

  One particular list tends to receive large numbers of messages
over short time periods (eg 50 messages with less than a minute)
followed by long periods of inactivity (eg an hour).  The result is
that multiple (typically a dozen) simultaneous instances of mailman
are invoked for that list.  

  Watching a mail bomb come thru the apparent behavious is curious:
The first batch gets thru fine (ie # of simultansous deliveries as
set in the MTS cfg), each individual message seperated from the
others by fractions of a second.  But the second run (which all
start within milliseconds of each other) all seem to block on
<listname>.lock for the same time period, and to then spin, all
unlocking and relocking in synch and getting nowhere.  Certainly the
wait time on the lock is rather long (I haven't checked the source).

Perhaps a shorter default lock time plus a random factor?

Eeek.  System is a 400MHz PII, 128Meg RAM, Python 1.5, RH 5.2, all
updates, latest CVS version of MailMan.

-- 
J C Lawrence                                   Home: claw@kanga.nu
---------(*)                Linux/IA64 - Work: claw@varesearch.com
 ... Beware of cromagnons wearing chewing gum and palm pilots ...