On Sat, 24 Apr 1999 10:35:07 EDT, "K.V. Rao" <[log in to unmask]> said:
> I have made the following suggested change and the change to sendmail.cf
> file. The result of this change seems to have worsen the situation. I am
> even unable to login to the listserv running system after posting a digest
> to list with about 11000 members. I am running this on solaris 2.7, 160 MB
> memory, Sparc 20, lsv 1.8d. This system is exclusively used for listserv
> and a small web server. The system loads have shot up to 12, 14, and to
> 16. I have not seen such high loads before. Any other tips? Thanks
1) As far as I can tell, kvrao.org is *not* listed in intpeers.names,
and as such, may not be taking full advantage of the Listserv
DISTRIBUTE protocol. This would mean you'd have to handle all 11K
recipients by yourself.
2) Sendmail will fork off another process every MAXBSMTP recipients,
or possibly every 40 or so (depending whether Listserv chops it
off after MAXBSMTP, or if Sendmail fills a buffer internally first).
Locally, we run with MAXBSMTP=250, and a Sendmail mod to match
(basically, raising the values of 'MAXLINE' and 'TOBUFSIZE' to
each be 4096 in sendmail/src/conf.h). This results in fewer, but
longer running processes. Also, you *might* want to look at
the value of SORT_RECIPIENTS in go.user.
3) 160M might be just a tad too little memory - check if you're
thrashing. Listserv.vt.edu has 256M, which lets all the Sendmails
run together, and still have room for /var/spool/mqueue to live
in the cache.
4) Look at setting CheckpointInterval in sendmail.cf - the default
is after every 10 deliveries. I run with it set to 100 on listserv.vt.edu,
saving about 90% of the disk I/O from checkpointing.
5) Look at using sendmail/contrib/re-queue.pl and running multiple
Sendmail queues.
6) Sendmail by default runs with a lot of very generous timeouts.
To keep everything moving, I reset them all to the RFC1123-mandated
minimum values.
7) If you're seeing TOO many sendmails running at a time and causing
thrashing, you *might* want to look at the sendmail.cf MaxDaemonChildren
setting. Be careful here, it *is* possible to throttle your system
too far with this - you have to find the largest number of sendmails
that can all run at once. Setting it too low will result in not
enough sendmails running, and a queue that keeps growing.
I think I've put most of this, plus some other tuning tricks,
on http://black-ice.cc.vt.edu/~valdis/sendmail.html
/Valdis
|
