data on what makes web access slow [Fwd: end user impact of TCP speed (was: Re: fast tcp && win2000)]

From: Philipp Hoschka (ph@w3.org)
Date: Fri, Nov 26 1999


Message-ID: <383EB7ED.DB9A3FD1@w3.org>
Date: Fri, 26 Nov 1999 11:40:13 -0500
From: Philipp Hoschka <ph@w3.org>
To: www-wca@w3.org
Subject: data on what makes web access slow [Fwd: end user impact of TCP speed   (was: Re: fast tcp && win2000)]


 

attached mail follows:


Message-Id: <4.1.19991123134857.00972c10@mailee.research.telcordia.com>
Date: Tue, 23 Nov 1999 13:57:16 -0500
To: Joe Touch <touch@isi.edu>, ph@w3.org, chase@cs.duke.edu, gray@microsoft.com
From: Christian Huitema <huitema@research.telcordia.com>
Subject: RE: end user impact of TCP speed (was: Re: fast tcp && win2000)
Cc: sandrof@microsoft.com, feldy@myri.com, end2end-interest@isi.edu, gallatin@duke.cs.duke.edu, grant@duke.cs.duke.edu, SABAMA@exchange.microsoft.com, jawadk@exchange.microsoft.com

I have been conducting a little experiment to try assess precisely that --
are the long web delays due to network performance, or are they due to
server overload?

One way to get a cue is to look at the delay between the sending of the
HTTP/GET command and the arrival of the first byte of the response. In
theory, most of the delays here will be due to the server, which has to
dequeue the connection (accept), get to read the command (select, recv),
then process the command, format the response and send it. I know there are
obviously limits to such evaluations...

Anyhow, you can get a look at the early results at 
	ftp://ftp.telcordia.com/pub/huitema/stats/quality_today.html
The "delays between GET and first byte" that I observed represent between
30 and 40% of the duration of the average transaction. To ease that, you
probably need more powerful servers. Getting faster connections would
obviously help the other 60% of the delay.
-- Christian Huitema