netdev
[Top] [All Lists]

Re: Diffserv and NetApplications

To: SVR Anand <anand@xxxxxxxxxxxxxxxxx>
Subject: Re: Diffserv and NetApplications
From: Chris Wedgwood <cw@xxxxxxxx>
Date: Tue, 17 Jul 2001 01:31:00 +1200
Cc: netdev@xxxxxxxxxxx
In-reply-to: <200107161318.SAA17536@xxxxxxxxxxxxxxxxx>
References: <200107161318.SAA17536@xxxxxxxxxxxxxxxxx>
Sender: owner-netdev@xxxxxxxxxxx
User-agent: Mutt/1.3.18i
On Mon, Jul 16, 2001 at 06:48:07PM +0530, SVR Anand wrote:

    Thanks for your reply. I would certainly be happy to visit the
    sites that have tcp/udp applications using diffserv QoS
    mechanism. Can you please pass it on ?  I searched on Google
    without much success :(

No idea, I just noticed the QoS bits are sometimes set when using
tcpdump for telnet and ssh.  I don't think any special version exist,
just grab the source and take as look.

    With respect to your other point of the practicality of the
    proposed QoS, all I have to say is that it is atleast good to have
    QoS provisioning rather than not having at all.

But this means different things to different people.  I know of many
uses for QoS bits, that aren't strictly related to QoS issues (such as
tagging traffic on ingress to specific parts on an AS, for internal
metering purposes).

    At least in one case, I can clearly see advantage of adaptively
    using diffserv dscp values in the http servers to serve the urls
    in an intelligent and responsive way to its clients.

And the upstream provider and go and changes these value on you,
ignoring your QoS requests :)

    What is the big point of having architecture, and no one to use
    it?

QoS has great intentions, they just never really happenned for lack of
good core support, which we are only just beginning to see appear now.

    Don't you think we have to make a beginning somewhere?

Indeed, but I think QoS has almost missed the boat completely.  With
the rise of MPLS (be that good or bad), QoS might be finally become
something useful, or die completely.

Core routers now support LSP selection and even in the non-MPLS case
queue selection (for use with WRED) based on TOS bits.  (Actually,
this is a lie, Juniper support this, I don't think IOS does yet, and
if it does, I doubt it will at line-speed, but still, thats the
majority of the  emerginghigh-end anyhow).




  --cw


<Prev in Thread] Current Thread [Next in Thread>