[plt-scheme] Webserver on Vista, increasing system load

Perhaps your benchmarking utility is using HTTP/1.1 and not closing
the connections after the requests are done. The webserver will
eventually time out these connections, but it's a tuning parameter.
Those connections have a system-level existence, so that could have
something to do with the behaviour you're seeing.

Jay

On Tue, May 27, 2008 at 5:41 AM, JoopR <joop.ringelberg at gmail.com> wrote:
> Hi,
> I have tested the webserver on Vista Business using JBlitz pro (a
> webserver stress test utility). The test requests a page generated by
> a Scheme servlet that retrieves an Xml fragment from the eXist
> database. The page contains a url that is used by the stress tester to
> request the next page. This happens 7 times in sequence (7 pages) and
> then the test case restarts with the original URL.
> I've tested using 20, 40, 80 and 120 virtual users.
> I see the same pattern in all cases:
> * the webserver gradually increases its memory use;
> * response time gradually increases from around 100 msec to 4 - 5
> seconds (after 40 minutes of testing)
> * the "System" process gradually claims more time from the processors,
> from about 1% up to 50 - 60%
>
> I ran a Perl script that generates exactly the same page, ActiveState
> Perl on IIS. It also does the same interaction on the eXist database.
> Ran the same JBlitz test script. Response times are similar to the
> initial times of the Scheme version, but do not rise during the test
> as in Scheme. This seems to rule out that the eXist component is to
> blame (memory consumption by eXist is pretty much the same during the
> testrun, after the maximum allowed has been reached).
>
> However, the System activities persist after the PLT webserver has
> been shut down, causing the Perl response times to be an order of
> magnitude higher (when run after PLT has finished). It looks like some
> kind of resource is initially claimed by the PLT webserver, and then
> is not released, giving the System process a lot of work (whatever it
> does).
>
> I lowered the various timeouts in the PLT configuration, with overall
> better performance, but no effect on the decline described above.
>
> My questions:
> * Any ideas on the cause of this behaviour, and a possible cure?
> * Does the same happen under Unix?
>
> Joop Ringelberg
> _________________________________________________
>  For list-related administrative tasks:
>  http://list.cs.brown.edu/mailman/listinfo/plt-scheme
>



-- 
Jay McCarthy <jay.mccarthy at gmail.com>
http://jay.teammccarthy.org