httpd-test-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "David Hill" <ddh...@zk3.dec.com>
Subject Flood output analyzer...
Date Wed, 15 Jan 2003 22:08:14 GMT

Hi all,
    I been playing with flood a bit.

One thing that I liked in ab that I found missing in flood was the
statistical summary that ab does.

While analyze-relative does the average stuff, I wanted the standard
deviation and the percentile ramp. As I didn't see anything that did
that, I whipped one up that I called flood_stat_report. While it could
use a bit of internal improvement I thought I would run it up the flag
pole and see what others thought. If anyone else thinks it useful,
perhaps we could stuff it into the tree so the next me will not have
to rewrite it :-). I will attach it to this email.

A question about distributed load generation..... has anyone thought
much about how to look at the load when it stabilizes ? It would seem
to me that you want to measure the load at steady state, and that
there is a window after all of the flood clients start and before the
clients start to finish that would be the best measure of the reponses
of the system. One approach would be to combine all of the client
outputs  and lop off the beginning and end results - then look hard at
the middle bits. There are a few ways to get at the middle bits
(looking at all the files to combine and selecting only the common
ranges, or perhaps running for x + 2 minutes and then lopping the
first and last minute off). Has anyone tooled anything like this ? My
program will do the skip time + window thing, but kind of assumes that
the input data is sorted (it uses the first entry to pick the starting
absolute time which might not be very acurrate if I just cat-ed a
number of results files together).

Dave.

Mime
View raw message