Re: [tcpm] draft-eggert-tcpm-historicize-00

[Hagen Paul Pfeifer <[e-mail]>]

* [e-mail] | 2010-06-09 21:13:34 [+0100]:

>I've read such papers. They tend not to discuss the initial SYN/ACK handshake
>and the timer related to that (probably because their simulation tools default
>to not simulating SYN/ACK), or consider the implications of TCP slow start with
>long propagation delays.

It is a little bit off-topic, but can you comment on the actual problems of
the initial SYN/ACK handshake and the timer interplay? Linux for example use
the same RTO timer for the initial handshake and the ongoing connection. If
you tune the RTO timer you will also tune it for the initial three way
handshake (this statement is not 100% sure, I hide some petite differences,
but the effect is insignificant). Why differs the handshake so much in your
eyes? And yes, don't trust any simulation models that you have fully
developed by your own. ;-)

>the sensor networks I'm familiar with use IP, but don't use TCP, for
>many of the same underlying reasons.
>http://sat-net.com/L.Wood/dtn/saratoga

Yes, the characteristics and problems are nearly the same. The papers I read
("Making TCP/IP Viable for Wireless Sensor Networks", "Hop-by-Hop TCP for
Sensor Networks", ...) are pure academic papers and I personally saw no sensor
network TCP implementation in the wild.


Hagen