Hi!
As summarised in section 2.5, the auto-tuning technique adjusts the buffer
size based on the utilised window. This entails increased advertised window
until packets are lost. This thus contradicts section 2.3 that recommends
the advertised window to be limited to avoid buffer overflow in the queue to
the bottleneck link. We believe the latter recommendation to be valid for
slow links.
We are however much in favour of TCP auto-tuning since the bw*delay product
could be considerable over some wireless links and a mobile node could roam
between different access links. We would thus like the auto-tuning section
to remain somewhere, somehow in the output from PILC but it is not needed
for the slow links.
Sorry for the late comment.
Haider Shakargi & Jan Kullander
PS
It would be nice to combine auto-tuning with an optional max window
associated with interfaces...
DS
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 09:12:24 EST