Spencer,

Next time the web interface hangs, check your spool directory and see
if there are a large number of jobs being processed. The way I understand it,
and someone correct me if I'm wrong, is that Listserv is single-threaded so
if it's busy processing a lage email job, the web interface is going to hang
until it finishes the email job. We used to have that exact same problem,
especially when lists over a few thousand people started replying like crazy,
and one of the things we did to help it was add a SMTP_FORWARD_1= line to the
go.user file. Since then the mail is passed off of the listserv machine
quicker, and the web interface rarely hangs anymore.

-Jim

--
Jim Serwinowski                 [log in to unmask]
UB Listserv Administrator       http://listserv.acsu.buffalo.edu


On Fri, 14 Jun 2002, Warhurst, SI (Spencer) wrote:

> Date: Fri, 14 Jun 2002 17:11:01 +0100
> From: "Warhurst, SI (Spencer)" <[log in to unmask]>
> Reply-To: LISTSERV give-and-take forum <[log in to unmask]>
> To: [log in to unmask]
> Subject: Web interface performance
>
> Hi
>
> Our main JISCmail service currently has about 4400 lists on it. It is a
> standalone server, running Windows 2000 Server (recently upgraded from NT4)
> and Apache 2 (recently upgraded from 1.3x) on a dual-Xeon 550, with 512Mb
> RAM and a hardware RAID5 disk config.
>
> I am becoming increasingly concerned about the performance of the web
> interface.
>
> At peaks times the web interface has always slowed down a bit, but maybe 1-3
> times a week it would grind to a halt altogether. At these times, when I
> looked at the server, I would find that there were 50 wa.exe processes
> running (Presumably 50 instances of a process is an NT/W2K or Apache
> limit?).. Now, the processor usage would be fine (say not more than 40-50%)
> and there would be 100-250Mb of free physical memory. Disk space is not a
> problem. What's more, while the web interface had ground to a halt, other
> pages not requiring CGI were perfectly navigable. In other words it only
> seemed to be the CGI locked up.
>
> I thought what maybe happening was that as the web interface got a bit
> slower at peak time, users would start a process, get impatient, hit their
> browser's Stop button, but the process on the server would still be active.
> Then they'd try again and cause another wa.exe process to start. Then a
> couple of dozen other users would do the same thing and before you know a
> jam of processes occured. On some ocassions, to get the web interface going
> again, I had to use the NT resource kit's Kill utility to kill all the
> wa.exe processes. Not an ideal course of action but it got the web interface
> generally working again!
>
> However, I am still no wiser as to what the *cause* of the problem is and
> why it only seems to be affecting CGI. I was rather hoping that upgrading to
> Win2k/Apache 2 would help solve the problem. While, in the last 2-3 weeks
> since the upgrade, we haven't had a case where the processes have built up
> to 50, the web interface still can get very slow to the point where it takes
> 5+ minutes to work, or times out.
>
> We also want to upgrade Listserv to 1.8e, but I am concerned the new
> features will exacerbate the problem. Has anyone else here had similar
> problems? How do you find performance of the web interface on your servers,
> how many lists do you have, what hardware spec etc?
>
> Thanks
>
> --
> Spencer Warhurst
> Listserv Service Manager
>
> JISCmail -> http://www.jiscmail.ac.uk
> Mailtalk -> http://www.mailtalk.ac.uk
>
>
>