[5561] in www-talk@info.cern.ch
Re: holding connections open: a modest proposal
daemon@ATHENA.MIT.EDU (John Ludeman)
Tue Sep 13 20:19:06 1994
Date: Wed, 14 Sep 1994 02:12:15 +0200
Errors-To: listmaster@www0.cern.ch
Errors-To: listmaster@www0.cern.ch
Reply-To: johnl@microsoft.com
From: John Ludeman <johnl@microsoft.com>
To: Multiple recipients of list <www-talk@www0.cern.ch>
----------
| From: Gavin Nicol <gtn@ebt.com>
| To: Multiple recipients of list <www-talk@www0.cern.ch>
| Subject: Re: holding connections open: a modest proposal
| Date: Tuesday, September 13, 1994 5:22PM
|
| >Briefly, keeping the connection open requires the server to commit
| >resources until timeout or explicit quit, reducing the number of
| >connections that can be made at one time and reducing the throughput.
|
| This is true, one can architecture servers such that things like file
| handle limits do not apply, but there is still an increase in CPU
| usage when connections are left open. I have never measured the
| different loads, but I would guess that the overhead of having open
| connections, and the overhead of multiplt connections/terminations is
| much the same.
No, you pay a lot for session setup and tear down. You pay in CPU
cycles and resources on the server. You pay in wasted network
throughput for the client which is *extremely* costly on slow links
(aka modem) especially with the ever growing set of hypermedia
documents that have lots of little graphics.
A server that is taking up a large portion of CPU on a non-active
connected socket is a server that should be replaced.
John
