From: Hancock, David (DHANCOCK) <DHANCOCK@arinc.com> - 2002-09-12 15:24:42
Nick: I have heard of something called the "global interpreter lock" in
Python that sounds like it means only one process can get hold of some
critical part of the interpreter at one time. Maybe that's related, maybe
I'm uninformed... Probably the latter.
As nice as httplib is for benchmarking, though, I tend instead to use 'ab'
(the benchmarker that comes with Apache). This lets you specify the number
of concurrent accesses, the total number of accesses, and the URL (among
other things), and it runs and spits back a lot of statistics. I've used it
for benchmarking WebKit with and without mod_webkit, with 100 concurrent
accesses and 10,000 total requests, and never got a core dump.
Here's a quick command line:
/usr/local/apache/bin/ab -c 20 -n 2000
You can pass ab basic authentication, custom headers, POST data, etc. It
seems fast enough that its own overhead doesn't skew the statistics.
David Hancock | dhancock@... | 410-266-4384
From: Nick Murtagh [mailto:nickm@...]
Sent: Thursday, September 12, 2002 10:54 AM
Subject: [Webware-discuss] Python segfaults under load
I'm having a problem with Webware 0.7 using mod_webkit,
python 2.0 / 2.1, apache 1.3.22 on a Mandrake 8.0-ish
I've written a simple benchmarking script that uses httplib
to retrieve a page from the server inside an infinite while loop. This works
fine. If I run another copy in parallel on the same machine, after a while
all the python processes segfault and coredump. I'm in the process of
getting together a version of python with debugging symbols compiled in so I
can get some information out of gdb, but until then, has anyone seen
anything like this using Webware?
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
Webware-discuss mailing list Webware-discuss@...