[e2e] To make my thought more clear.

[Detlef Bosau]

In the traditional way of VJCC, we often think packets were distributed
along the path quite equally.

Now, imagine

sender --------------- Internet cloud --- GPRS link --------- Receiver

and a 1 MByte CWMD.

What happens when the GPRS link suffers from a "bandwidth shortage"?
The whole 1 MByte CWND piles up at the router before the GPRS link.
So, while the "Latency Bandwidth Product" may well be 1 MByte, when this
amount of data is in transit and equally distributed along the path,
exactly this amount of data may cause severe congestion when gathering
at one interface.

This could not even be alleviated by adding buffer: Our probing schemes
will have this buffer fully utilized, which offers two advantages:
1. The end to end latency increases.
2. CWND increases as well - and the next congestion problem is to be
expected ;-)

In Reiner Ludwig's PhD thesis, he talked about "hiccups" (spelling by
Reiner Ludwig).

These and perhaps extreme throughput steps from, say, 10 GBit/s down to
100 kBit/s may cause data to gather at individual interfaces along the
path which renders our "Little's Law model" (this is exactly the idea
behind our abstraction) somewhat questionable.

(If you prefer a wireline example: A "transient bandwidth shortage" may
be caused e.g. by cross traffices along the path.

Detlef


-- 
------------------------------------------------------------------
Detlef Bosau
Galileistraße 30   
70565 Stuttgart                            Tel.:   +49 711 5208031
                                           mobile: +49 172 6819937
                                           skype:     detlef.bosau
                                           ICQ:          566129673
detlef.bosau at web.de                     http://www.detlef-bosau.de