To: [email protected]
Subject: WWW usage
Errors-To: [email protected]
Reply-To: [email protected]
Organization: Berkeley Software Design, Inc.
Date: Fri, 25 Jun 1993 23:23:39 -0500
From: Tony Sanders <[email protected]>

 FYI: This document_ is available via the setext -> HTML converter:
   http://www.bsdi.com/setext/www-usage.etx

WWW Usage Tracking
------------------
 Is anyone interested in tracking WWW usage?

 I propose someone setup a mail gateway for usage reports that get dumped
 into the Web.  I would be willing to do this if no one else is interested.
 I could at least archive the raw data onto the Web if not summerize it and
 graph it.

 Maybe a weekly report that says something like:

   Date: Sat Jun 26 03:01:56 GMT 1993
   Server: http://www.bsdi.com/
   Requests last 7 days: 3589
   Unique hosts last 7 days: 69
   Total unique hosts: 847
   Total requests: 18358
   Total days of service: 52

 Whatdaya think?

 At some point it would be neat if someone could take all the IP addresses
 from all the servers and ``uniq'' them to get an idea of how many users
 we have (My server has 847 unique ip addresses in the log).  Does anyone
 care about having their IP address uniqued?  Does anyone care about having
 their IP address made public (I'm not suggesting this, just curious).

 Once HTTP/1.0 gets more widespread we will be able to track client usage.

 In case anyone else is interested: my 7 day summaries for the past
 52 days (minus this week).  Each * is 100 requests.

   05/03 ********************
   05/10 ********************
   05/17 ******************
   05/24 ***************
   05/31 ***********************
   06/07 *****************************
   06/14 *******************************

 Anyone else have this data they can post?

A Related Issue
---------------

 Is anyone having any performance problems with their HTTP servers?  I did
 some tests on my 486DX2/66 (call it 31MIPS) and found that Plexus runs
 about 5 tps, and I timed the NCSA server that runs from inetd but it's
 not a fair comparison (1 tps) because you could run it native and save
 the exec cost.  What are you busy server people running at peak (in tps)?
 I would **really** like to hear what tps other people are getting and on
 what hardware (~ MIPS anyway) and if you can do some peak performance
 tests that would help.  I expect it'll be about the same (5 tps peak) on
 most hardware available today as I think most of the overhead is beyond
 the control of the server (i.e., in TCP/IP and exec).

 --sanders_

. _document http://www.bsdi.com/setext/www-usage.etx
. _sanders http://www.bsdi.com/hyplan/sanders.html