W3C home > Mailing lists > Public > www-talk@w3.org > September to October 1996

Re: study completed

From: Gwenael Durand <gd@antares.ibp.fr>
Date: Thu, 26 Sep 96 14:17:16 +0100
Message-Id: <9609261317.AA00551@antares.ibp.fr>
To: www-talk@w3.org
Well, I first believe that statistics are obviously uncompatible with caching.

Second most load on servers comes from accepting a new connection, forking a 
new process to serve it, scanning the file system to find the requested file,
but NOT ALWAYS FROM TRANSFERING files which seam to be smaller day to day.

Based on your solution, an idea could be to force the client or cache to send
an If-modified-since GET request, to logg it, and then to answer no. This would
first save the cost of looking throught the file system. Your server could 
also be optimized for servir if_modified_since requests without starting a new
process each time.

The most critical point of these solution is to force caches to send an 
If-modified-since request. Assigning a small enouth time-to-live (or big enouth
last-modified) to one of your inline images, will not be enought to make you sure you will log all actual requests, but it tells the caches how many time
they can keep the image without trying to refresh it. This would be sufficient
enought for not so popular pages.

 Gwenael Durand
/\_\ Labo MASI - P. et M. Curie - P6
\/_/ gd@sirius.ibp.fr
Received on Thursday, 26 September 1996 08:50:48 UTC

This archive was generated by hypermail 2.3.1 : Tuesday, 6 January 2015 21:32:59 UTC