[e2e] Number of persistent connections per HTTP server?

[Micah Beck]

Interesting - kind of a little bonus to proxies as a reward for violating
end-to-end.

Seriously though folks, this raises an issue as to whether overlay networks,
multicast infrastructures and other mechanisms that deal with aggregate
communication on behalf of some community should be allowed to use network
resources in a way that would be considered unfair if a single-user
application behaved that way.

Content distribution networks can push data out to the edge in order to
optimize aggregate traffic; should they be allowed to use a higher number of
concurrent TCP streams?  If so, what formula should they apply in order to
be responsible?  Must it be based on the number of end-users currently
consuming the data (as in a proxy) or can it be based on some estimate of
the number of users to be served over time?

This raises a real possibility: if users of multicast, including overlay
multicast, were to be given a substantial bonus in terms of bandwidth
available to them in aggregate, this might give them the motivation to, for
instance, batch up large transfers over long distances.  However, policing
such a scheme seems almost impossible.

Micah Beck
University of Tennessee

----- Original Message -----
From: <Jim.Gettys at hp.com>
To: "Spencer Dawkins" <sdawkins at cynetanetworks.com>
Cc: <end2end-interest at postel.org>
Sent: Sunday, October 13, 2002 10:25 AM
Subject: Re: [e2e] Number of persistent connections per HTTP server?


>
> >
> > Mozilla seems to be implementing features like HTTP request pipelining,
so
> > this didn't surprise us a lot - we're just curious what science is
behind the
> > distinction between "number of connections per server" and "number of
> > connections per proxy".
> >
> > Any pointers?
>
> The working group archives.
>
> There is a fairness issue: a proxy is usually acting on behalf of many
> users.  So it should be able to get correspondingly more bandwidth out of
> the network, in proportion to the number of users of the proxy.  So if
> 10 users are accessing a web site through a proxy, it ought to be allowed
> to get 10 times the bandwidth of an individual.
>
> >
> > Spencer
> >
> > (And, as an aside - the last posting on the HTTP mailing list was in
> > September of last year - is there a more appropriate place to ask
conceptual
> > questions about HTTP mapping to TCP connections?)
> >
> The mailing list moved last year: it is still active.
>                             - Jim
>
>