On Wed, 13 Nov 2002 10:30:18 CST, "Walker, Brand" <[log in to unmask]>  said:

> We have approximately 700 lists, and periodically daytime loads spike to 2.42
> or as high as 4.67.  The /home/listserv/spool directory fills with a backlog
> of .job files.  While lsv continues working, the web interface hangs out and
> then fails as the browser times out.

The problem is that although much of the sendmail/apache stuff will multithread
quite nicely, the 'lsv' process itself is still mostly single-threaded.  As
a result, if the wa.cgi tries to wake lsv up while it's busy, it has to wait
until lsv is ready to answer - by which time a timeout may happen.

There are 2 major bottlenecks that you can get bit by:

1) It can take a long time for lsv to hand mail to sendmail, especially with
a large number of recipients.  Things that help here:  (a) use the
FEATURE(nocanonify) on the Sendmail side (which will get you literally a 5-8x
speedup as it cuts out a DNS lookup at an inopportune time) and (b) on the
lsv side, coding something like this in the go.user file:

SMTP_FORWARD="localhost"
SMTP_FORWARD_1="3*localhost"

This will cause lsv to fork off 3 child processes that will do the actual
handoff to sendmail - and the parent lsv can start the next request as soon
as a child *starts* to do the work rather than waiting for it to finish.

2) A large 'ADD' job to bulk-load a list.  Only real fix here is to cough
up the money for a Listserv HPO license.
--
                                Valdis Kletnieks
                                Computer Systems Senior Engineer
                                Virginia Tech