Images and CSS generating a lot of bandwidth

Gerard Beekmans gerard at linuxfromscratch.org
Fri Sep 19 14:31:57 PDT 2003


Hi guys,

I'm sure by now most of you know that I've been working on ways to
minimize our bandwidth usage. We can use up to 2.5 GB data a day. Lately
(this month and most of last) it's been closer to 2.8 to 3 GB a day.
Every now and then this isn't an issue, as long as there are days where
it's less than 2.5 GB to even things out. We're getting too close to our
limit to be comfortable for me.

I've been keeping an eye out on things today using iptables byte
counters and webalizer analysis. If webalizer is correct, we've used 950
MB today in HTTP traffic (it's 3:30 PM now). Over 200 MB of this is
images and stylesheet downloads.

We (me and mirror admins) are discussing using round robin DNS but there
are issues with that too. If are sent from random mirror to random
mirror every time you click on a link, we'd better make sure all mirrors
that are in the random pool are identical to the master server. Imagine
if you are reading a CVS book. There's a good chance you're reading
today's CVS when install binutils and yesterday's CVS when installing
the next package GCC. This can lead to all kinds of interesting bug
reports (that later are written off as mirror bugs -- very annoying).

So, I'm wondering if the images and stylesheets can't be shrunk in size?

I'm also exploring other avenues such as getting mod_gzip working in
Apache which will compress data with gzip before it's sent to the client
(provided the client supports this). The problem I'm facing is that we
run Apache-2 and I haven't found a mod_gzip that works on Apache-2 yet.
Maybe I'll downgrade to 1.x if that's going to be the only solution.

I alluded to the point that webalizer might not be accurately reporting
used bandwidth. At midnight when the logs are rotated, the iptables
counters will be reset as well. Starting tomorrow I can check at any
time how much data webalizer says has gone through http and how much the
kernel firewall has logged.

Another option was to remove the master server from the mirror list. I'm
not sure how much that will matter in terms of bandwidth. There are
enough people who use direct links, bookmarks, google results, etc.



-- 
Gerard Beekmans
http://www.linuxfromscratch.org
http://www.beekmansworld.com

/* Linux Consultant --- OSDN / DevChannel *
 * Technical Writer --- CheapBytes        */

/* If Linux doesn't have the solution, you have the wrong problem */





More information about the website mailing list