> -----Original Message-----
> From: Eric Dean [SMTP:edean@gip.net]
> Sent: Monday, May 11, 1998 12:57 PM
> To: Constantinos
> Cc: end2end-interest@ISI.EDU; ippm@advanced.org
> Subject: Re: Current use of RED?
>
>
> We experimented with RED a while back and found that most of our flows
>
> are non-contiguous packets. As a result, whether we used
> tail-dropping
> methods or RED methods or WFQ, we saw little result within a highly
> concentrated backbone upon end-to-end session performance.
>
> Our flows did not consist of back-to-back packets (which is
> why IP compression didn't work either). In addition, we find that
> pretty
> much all of our flows always have a slow-start window size as well and
> only
> consist of about 10 packets per flow (WWW flows = 70% of our network
> traffic).
>
> We toyed with the idea of using a Queueing method for Class of
> Service,
> but then got stimyed by the Manageability and Scaling requirements.
>
>
> On Mon, 11 May 1998, Constantinos wrote:
>
> > A major question for anybody that plays around with traffic traces
> these
> > days is if the routers in the major backbones already use RED or
> not.
> > There is no way that you can explain what you measure if you don't
> > know how packets are dropped.
> >
> > I was wondering if the large backbone providers would like to share
> > with the end2end-interest and the ippm working groups the extent in
> > which they use RED (or variations), and any early experiences they
> have
> > got deploying it.
> >
> > I hope that such information will not be considered "proprietary".
> >
> > Thank you so much,
> >
> > Constantinos
> >
>
> _______________________________________________________________
> Eric Casey Dean
> Supervisor: IP Product Engineering
> Tel#: 703-689-5298 Fax#: 703-478-7852 Mobile#: 703-598-0962