Reiner:
>> From rludwig@huginn.cs.berkeley.edu Wed Jan 27 16:29:01 1999
>>
>> The paper that Venkat refers to ("Multi-layer Tracing of TCP over a
>> Reliable Wireless Link") can be found at
>> http://iceberg.cs.berkeley.edu/publications.html.
I found it interesting that (in the above paper)
you did not see many competing retransmissions in
the study of TCP over GSM - that is, TCP did not
timeout despite link level (RLP) retransmits.
Do you know if this conclusion is valid if you restrict
the window to at most, say, 2 or 4 segments ?
Small window sizes should be enough to fill the pipe.
Since the TCP window in your measurements tends to get large,
that leads to large RTO, as you also point out. This, in conjuction
with the 4*rttvar added to RTO, may be making the RTO large
enough that the TCP timer does not go off despite RLP retransmits.
Seems to me that at small window sizes competing retransmits
would become an issue.
(By a similar argument, the impact of RLP link resets
may become smaller at smaller window sizes, as
compared to what you observe.)
Am I missing something here ?
- nitin
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 09:12:20 EST