httpd-test-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jacek Prucia <>
Subject Re: flood: responsescript patch
Date Wed, 13 Aug 2003 09:35:02 GMT
On Mon, 11 Aug 2003 16:26:34 -0700
Justin Erenkrantz <> wrote:

> --On Sunday, August 10, 2003 23:24:04 +0200 Jacek Prucia 
> <> wrote:
> > This probably belongs in contrib/patches. It is a quick'n'dirty hack
> > I did few days ago, to simulate applet making network connection.
> > Basically it allows for something like this:
> >
> > <url
> > responsescript="/path/to/"></url>
> I wouldn't see an issue just placing this in the main code rather than
> as a separate patch.  Just place in the documentation that the use of
> this attribute may distort the timings.

OK. Will do after the rewrite (see Aaron post).

> Hmm.  If we did this *after* the close is done, would we still have a 
> problem with the timing?  I don't think we would, but I could be
> wrong. (Perhaps we're freeing the response buffer at that time.  I
> forget.)

How about a keepalives? We keep socket open until "Connection:
Close" header arrives, or we run out of urls (IIRC).

However, this not the only place, that can bork timings. We will have
more of them in future, as flood can be used in two types of tests --
performance and regression (with capacity planning being a mixture of
the two). The first one tries to produce as many requests as it can, and
pays more attenion to timings. The other is basically only interested in
response and it's content. So as development goes on with
regression-type features (my personal puppy), we will have more time

There are two ways to reduce that. The first one is to have two modes of
operation with magic attribute: dynamic. For example, variable expansion
in urllist would only occur, if that url list would have attribute
dynamic set to "yes". Flood could ignore few other attributes (including
responsescript) when in non-dynamic mode. This would also allow us to
clean url element a bit, as it is bloated with attributes. 

The second way, is to have precise timing of interesting operations,
rather that time check at milestone flood phases (what we have right
now). Something like that (pseudocode):

time_start = time();
time_end = time();

exact_time = time_end - time_start;

As time operations seems to be expensive (IIRC), we might use such
precise timing where such precision is really needed (like response time
of web server) and standard measuring else. However, I'm not so sure if
this is the right way, as this (probably, not sure) introduces some
issues in how we report timings.

> Watch out for C++-style comments (i.e. //) - that's not supported by a
> bunch of C compilers.  ;-)  -- justin

No to mention violating ASF style guide. Thank's for catching this one
out. I do a lot of Javascript recently (Flood URL Bag for Mozilla), and
this is how that little thing sneak in :)

Jacek Prucia

View raw message