[e2e] Number of persistent connections per HTTP server?
pradnya at cc.gatech.edu
Tue Oct 22 08:36:36 PDT 2002
Thank you for your interest in our paper. Here are the answers to your
The paper focuses on defining and achieving multipoint-to-point
session fairness, assuming we have global knowledge of link
capacities. We are currently working on a distributed version of the
algorithms, with limited knowledge at the end-hosts.
You have made a very good point about short-lived connections. In this
paper, we target static sessions in which connections start and end at
approximately the same time, and which are relatively long-lived. We
plan to extend this framework to dynamic sessions, with connections
starting and terminating at arbitrary times.
One way to prioritize particular connections within a session would be
to give a higher weight to those connections. So in your example, the
client could set a higher weight to the connections for the conference
application, and a much lower weight for the advertisement connection.
Hope this helps. Let me know if you have any other questions.
Networking and Telecommunications Group,
College of Computing, Georgia Tech
On Sat, 19 Oct 2002 at 10:05am +0800, Jing Shen wrote:
> I've read Karbhari etc's paper "Multipoint-to-point
> session fairness", but I have some question with it.
> The two algo. proposed try to adjust source rate with
> regarding to intra-session weight and inter-session
> fairness. But,
> 1. how could that be done in internet?
> If it's to adjust rate between servers, it just need
> to consider e2e transmission capacity, but the algo.
> make use of link bandwidth info along the routing
> path. If it's to adjust IP packet enjecting rate
> between hosts, how could it guaranteed to figure out
> link capacity along the routing path and inform the
> exact value to server?
> some of the connection is short-lived!
> 2. the algo. seems to adopt a equal-treating
> methodology between all session/connection, but what
> about those demand with different priority? such as,
> if there is advertisement info on a web conference
> application, the participants engage a priority on
> conferencing packets other than those advertisement
> >We've recently been working on higher (session) level
> >of fairness, with a focus on a range of possible
> policies and
> >algorithms to achieve them. Our emphasis has been
> >single-receiver sessions, as arise for parallel
> complex web pages, etc. See:
> >Multipoint-to-point sesssion fairness in the
> >Karbhari, Zegura and Ammar. Submitted to Infocom'02.
> Jing Shen
> State Key Lab of CAD&CG
> ZheJiang University(YuQuan)
> HangZhou, ZheJiang Province 310027
> Do You Yahoo!?
More information about the end2end-interest