[e2e] end2end bandwidth estimation

Constantine Dovrolis dovrolis at cc.gatech.edu
Tue Jul 22 14:43:16 PDT 2003


Available bandwidth can also exist in a path, together with the two
reasons you mention, because of the limited size of a flow. In other
words, even if a TCP flow is not limited by the sender/receiver
socket buffer sizes, it may end before it gets to increase its
window to the point that it saturates the bandwidth in its path.
A very crude way to think about it is that if flows arrive to a link
with an average rate f flows/sec and the flow size is s bytes/flow,
then the offered load to that link is f*s bytes/sec.

I suggest you look at the paper "On the characteristics and origins
of Internet flow rates" by Zhang et al. in Sigcomm 2002. They
analyzed a bunch of traces to identify which factor limits each
TCP connection's throughput. Only a small fraction of flows is
limited by the congestion window (even though the same fraction in
terms of bytes is significant - 20-50%).

Another way to think about it is that available bandwidth exists
in most paths out there because ISPs loose customers if they
allow their links to be persistently congested.

Regarding applications of available bandwidth estimation, you can
see the last section of our Sigcomm02 paper. But it is only just a
brief list.


Constantinos Dovrolis | 218 GCATT | 404-385-4205
Assistant Professor | Networking and Telecommunications Group
College of Computing | Georgia Institute of Technology
dovrolis at cc.gatech.edu

On Mon, 21 Jul 2003, Vivek Rai wrote:

> Hi,
> 	The end2end bandwidth estimation problem is categorized in three main ways
> depending on its applications:
> 1. Bottleneck bandwidth estimation
> 2. Available bandwidth estimation
> 3. Bulk Transfer Capacity estimation
> 	 There will be a sizable "available bandwith" along a link only if the
> flows traversing it are inelastic or they are rate limited due to maximum
> window size limitation. Since the Internet mostly comprises of the elastic
> traffic therefore the first assumption seems to be improbable whereas second
> assumption suggests that the links along the path are underprovisioned. In
> any case i am not able to understand what are the possible applications of
> measuring the available bandwidth.
> Thanks,
> Vivek

More information about the end2end-interest mailing list