Performance profiling

From: Alex Rousskov <rousskov@dont-contact.us>
Date: Fri, 24 Oct 1997 10:38:57 -0500 (CDT)

Hi there,

        I have some nice performance measurements for Squid you may want
to check out. I am very interested in your _participation_ as well as
comments and ideas on what to add or change:

http://www.cs.ndsu.nodak.edu/~rousskov/research/cache/squid/profiling/

        All the stuff is based on a simple patch that measures
_per request_ network and disk I/O activities. The results are quite
interesting. The patch has even been able to detect a performance bug in
Squid (that you probably want to fix, see the page for a cure :) ).

        The patch was run on two servers in the Netherlands (SURFNet: top
level and leaf), two servers in Norway UNINETT (top level and leaf), and
and more than a week on SV (NLANR, USA). The results include in-depth
performance analizes of each proxy and side-by-side comparisons.

        The collection may be essential for those who plan to install or
upgrade their proxies. You can actually see what kind of hit rates,
delays, and bottlenecks one may get in different environments. It is also
useful for performance enhancement of future versions of Squid.

        We do need to expand the collection though. If you like what you
see, please consider participating in our experiment! All you need to do
is to apply a patch and collect a day worth of logs. The patch does not
increase your CPU load and such. I will take care of prompt processing of
the results. I am especially interested in large leaf and top level
proxies (I have none from Asia and Australia, for example)...

        I hope you will find this stuff useful and will participate.

Thank you,

Alex.

P.S. This is a work-in-progress. Please pardon my bugs...
Received on Fri Oct 24 1997 - 08:57:10 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:20 MST