[e2e] TCP Performance over WWAN
detlef.bosau at web.de
Mon Sep 5 13:47:00 PDT 2011
On 08/14/2011 06:43 PM, Jim Gettys wrote:
> So attempting to retransmit a packet makes sense.
> The problem is, how many times? And when do you give up and try
> something different? And eventually, you really must drop a packet (or
> mark with ECN) to slow down the transmitter).
Could we just hang on on this very question.
How many times shall a base station retransmit a packet? And when shall
the base station give up?
Basically, this as implications for fairness and stability as well.
Fairness, because in WWAN a base station may serve more than one
mobiles. Hence, there is immediately a problem of resource allocation in
downlink direction and vice versa a resource allocation and MAC problem
in uplink direction.
Stability, because a base station doing unlimited retransmission causes
network load on the physical link. Hence, a "limited" input, e.g. a
single packet, may cause an "unlimited" output, i.e. an unlimited number
of retransmissions. This may immediately violate the BIBO stability,
which is typically pursued in the Internet.
We've well heard creeds on this issue, a very prominent one is the
Berkeley SNOOP protocol.
However, there is hardly anything concrete on this issue. Most of the
statements I've read so far, stay astonishingly vague.
More information about the end2end-interest