I'd have to disagree here. I worked on both sides of
the puzzle here, first on Netscape Proxy server and then
on IE. The main thing is that hosting servers tend
to be resource bottleneck, either CPU, disk, DB access etc..
Adding compressing significantly adds work to the system.
On the proxy server, we tested options to compress
content on the fly and in the cache. Most of our
customers had proxies deployed at the firewall or
edge of the network, which meant that all traffic
was driven through it. As a result, these boxes
were extremely busy and customers wouldnt enable
Outside of proxies, IIS (or apache) were busy running
CGI, ASP, or now PHP. Adding compression actually
slowed things down, since the servers were much busier.
Even though its clear that from a mathematical perspective,
the overall system may have been faster, the user perception
of speed is much different. While the compression was
a good thing for the internet as a whole system, site operators
were concerened primarily with the users perception of site speed.
Our studies in IE showed that the main factor was latency,
not bandwidth, and compression really only helps bandwidth.
CSS, the rendering phase of the page display became a bigger
and bigger bottleneck (on the client side). Adding a decompress
phase just slowed things down in terms of latency, eg time until
the browser could initially draw the page, or time until the user
could click a link.
People didnt seem to mind that it took longer for the page to
completely finish loading , which compression would have definitely
helped, or for images to trickle in , which it wouldnt help anyway.
>From: Andy Armstrong [mailto:email@example.com]
>Sent: Thursday, March 15, 2001 16:21
>To: Damien Morton
>Subject: Re: HTTP serving compressed content (was Re: Perl "competitor"
>Curl raises $52M)
>Damien Morton wrote:
>> > mod_gzip works pretty transparently with Apache.
>> IIS caches the compressed content, how does mod_gzip work?
>It compresses on the fly every time, but doesn't seem to slow things
>down too much.
>Andy Armstrong, Tagish
This archive was generated by hypermail 2b29 : Fri Apr 27 2001 - 23:14:15 PDT