[11724] in www-talk@info.cern.ch
Re: Persistant connections and pipelining...
daemon@ATHENA.MIT.EDU (Yves Lafon)
Wed Oct 25 15:30:52 2000
Date: Wed, 25 Oct 2000 21:22:28 +0200 (MET DST)
From: Yves Lafon <ylafon@w3.org>
To: Reinier Post <rp@win.tue.nl>
Cc: Nic Ferrier <nferrier@tapsellferrier.co.uk>, www-talk@w3.org,
gerald@w3.org
In-Reply-To: <20001025104119.D11758@win.tue.nl>
Message-Id: <Pine.GSO.4.21.0010252113090.3058-100000@tarantula.inria.fr>
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; charset=ISO-8859-1
Resent-From: www-talk@w3.org
Content-Transfer-Encoding: quoted-printable
On Wed, 25 Oct 2000, Reinier Post wrote:
> About right for the client, perhaps - but for the server, which may
> have to deal with thousands of other clients at the same time?
> If requests are expensive, a few clients with persistent connections
> may occupy much of the server's capacity.
It really depends on the way the server handle persistent connection. For
server that has the knowledge of the whole server state, it is possible t=
o
implement a LRU list of idle clients and free some when required (like in
Jigsaw [1]). Of course if your configuration is not "big" enough, this LR=
U
list will be too small and the timeout too short to really take benefit o=
f
persistent connections.
I tested the timeout on a persistent connection on jigsaw.w3.org when the
load of the server is the highest and I got more than 20mn of idling
before the connection was needed by the server.
[1] http://www.w3.org/Jigsaw/
--=20
Yves Lafon - W3C / Jigsaw - XML Protocol - HTTP
"Baroula que barouleras, au ti=E9u toujou t'entourneras."
