I saw a discussion on the PHPLIB mailing list about load testing, so I
tried some load testing on my development machine for midgard.
The results are below. I think it would be interesting to load test on
different boxes and see what kind of results we get.
Note that I am testing on the same machine, but not using localhost, so
this is going through a "normal" socket (not a Unix socket). Bandwidth
capabilities do not cloud the issue here.
Testing box:
SparcStation 20
Dual 50MHz processors, no cache (hey, y'all, remember this is a Sparc so
a 50 is equivalent to a P-150 or P-200 or thereabouts)
128MB memory
Solaris 2.6
Midgard 1.2.5
Testing command:
<path-to>/ab -n 500 -c X -k <address>
where -n 500 is the total number of requests for the test
-c X represents concurrency (simultaneous requests)
-k activates keep-alive
I tried this on both a static page and an active page.
The static page got to 34 concurrent requests before failing.
The active page got to 33 concurrent requests before failing.
Static page ran around 8 requests/second.
Active page ran around 5 requests/second.
Interestingly enough, I had no paging problems, but my cpus were up to
100% usage. This would indicate that, while sufficient memory is a
necessary thing, midgard is really responsive to SMP, and would probably
benefit very well from load-balanced clustering.
Some interesting numbers would be:
How the amount of memory affects a Pentium/Linux box?
What differences do we see between single-processor and multi-processor?
(Anyone got a quad out there)
Differences between Celeron and PIII (0 or 128kb cache on Celeron vs.
512kb on PIII)?
Differences among the various Linux ports? (This would be an
OS overhead question, maybe?)
the cat
--
This is The Midgard Project's mailing list. For more information,
please visit the project's web site at http://www.midgard-project.org
To unsubscribe the list, send an empty email message to address
[EMAIL PROTECTED]
