httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chuck Murcko <ch...@telebase.com>
Subject Re: request-time tracking patch to apache_1.1b2 (fwd)
Date Sun, 09 Jun 1996 02:17:04 GMT
Brian Behlendorf liltingly intones:
> 
> 
> For the record, being able to track in a log how long it took to serve a 
> file would be a very good thing - we have people asking us how long it 
> takes the average suchnsuch page to be rendered or the large application 
> to be downloaded or whatever, so this would be valuable.  Makes sense 
> that this should simply be another function in mod_log_config.c.  The 
> poster from sprynet.com implied that something had to be added to the 
> core to determine the initial connection time - yet the status module 
> is able to show how long an active connection has been alive, so I'm 
> wondering if the core actually needs to be altered to support this.  Any 
> guesses?
> 
We had a few go rounds about this months ago. We only know time since the
last request started, and not the actual time at which the reply started.

The consensus (I had a 1.0 prototype running for a bit) was to put a
second time stamp into request_rec that would get filled in just before
the reply started going out.

This evolved out of my desire to get all the log times coming from
a common set of timestamps, so they'd always agree across access, error,
and optional logs. It was also pointed out to me that you have to
walk the request_rec string for chained requests. The reply start
time would allow things like the status module to keep running tabs
on average, minimum, and maximum times to service requests.

Or you could send the raw timing data off to some bigger, badder
data reduction engine on another machine, and really go wild on
the performance analysis.

I had a *very* simple model running, but I still have my notes from
the suggestions David, Mark, both Robs, and others made on this,
if anyone's interested.

Things get less meaningful if you also want to have a timestamp for
replies completing, because that duration is so dependent on the client's
connection to Apache (and whatever lies between the two).

chuck
Chuck Murcko	N2K Inc.	Wayne PA	chuck@telebase.com
And now, on a lighter note:
Brook's Law:
	Adding manpower to a late software project makes it later

Mime
View raw message