Mon, 30 Jun 1997 14:48:03 +0200
|
On Mon, 30 Jun 1997 13:04:16 +1000 David M
<[log in to unmask]> said:
>My list would send about 200-300 X 1,500-2,000 = 300,000-600,000
>messages per day if I could run it to full boost (which I cannot
>unfortunately cause sendmail can't handle the load and it just queues
>and queues...)
How would multiple SMTP workers speed up sendmail and make it not queue
the mail due to lack of resources? Multiple SMTP workers have three main
uses:
1. When you deliver your mail through a remote host because you either
don't have a MTA on your LISTSERV host or it is really not much of a
MTA. This is typical for NT or older VAXes, and for people running
LISTSERV at home using their ISP's MTA. The MTA is often not in the
same location and may have significant latency. This is what the
feature was developed for.
2. To circumvent various operating system limitations or bottlenecks if
you have a really large workload. Even within a local system, there is
a turnaround time for socket I/O. Even a fast disk has nonzero access
time, etc.
3. On SMP systems, to allow you to use more than one CPU to input your
mail into the queue. Actually, even with one worker you can use 2 CPUs
concurrently (one for the worker, one for the MTA), but this is a case
where you eventually need multiple workers to use arbitrarily large
numbers of CPUs. Not that anyone actually has this problem but this
should be mentioned for the sake of completeness.
To give you an idea, on PLUM.EASE.LSOFT.COM (3.5 million daily
deliveries) we originally used 3 SMTP workers because of #2. Currently we
are running LISTSERV on a 4-CPU system to make sure there is no
multiprocessing bug in the code and we increased the number of SMTP
workers to 5 because of #3. It would work with 2, although there would be
a small backlog if we only had 1. We have tested it with 30 workers and
it just increases system overhead.
Eric
|
|
|