IETF Logo Mail Archive

Re: types of traffic in tcp congest control

J Wu <jinw@comp.leeds.ac.uk> Mon, 13 May 2002 18:30 UTC

X-Authentication-Warning: cslin135.leeds.ac.uk: jinw owned process doing -bs
Date: Mon, 13 May 2002 19:30:06 +0100 (BST)
From: J Wu <jinw@comp.leeds.ac.uk>
X-X-Sender: <jinw@cslin135.leeds.ac.uk>
To: Joe Touch <touch@ISI.EDU>
Cc: =?ISO-8859-1?Q?=B4=F3=C0=EE?= <liy666@sina.com>, <tcp-impl@lerc.nasa.gov>
Subject: Re: types of traffic in tcp congest control
In-Reply-To: <3CDFF18C.6060807@isi.edu>
Message-ID: <Pine.LNX.4.33.0205131918000.30118-100000@cslin135.leeds.ac.uk>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: owner-tcp-impl@grc.nasa.gov
Precedence: bulk
Status: RO
Content-Length: 945
Lines: 35

On Mon, 13 May 2002, Joe Touch wrote:

>
> Only if there _are_ elephants. In the transatlantic case, a huge number
> of mice can swamp the connection - yes, a severe shortage. One solution
> is to "RED" the SYN packets of new connections, or limit how many
> connections can be established.

In this case, UDP "mice" will still congest the network. Also, at the
angel of core switch, how can it decide mice from elephant? When same TCP
is supporting different application, it consumes different amount of
bandwidth. Will it help to do congestion control by just simply limits the
numbers of TCP connections. Never say that some TCP connection is not
always active.

And, is it technically possible to trace the TCP numbers at core switch?
Cheers, Jin.

>
>
> See above; this congestion, by definition, takes place inside the core,
> not at the access.
>
> Joe
>
>
>
>
>

-- 
Jin Wu
School of Computing,
University of Leeds.
Tel: +44 113 2336806

v2.8.7 | Report a Bug | By Email