[e2e] An alternative algorithm of calculating RTO

[Joe Touch]

jaglee at peoplemail.com.cn wrote:
> The algorithm in a nutshell:
> (1) Calculate mean measured round-trip time and measured round-trip time variation: Suppose mrtt[0], mrtt[1], ...,  mrtt[N-1] are the measured round trip time of the N most recently acknowledged packets without re-sending,
> 
> let mmrtt <- (mrtt[0] + mrtt[1] + ... + mrtt[N-1]) / N;
> and 
> mrttvar <-  (abs(mrtt[0] - mmrtt) + abs(mrtt[1] - mmrtt) + ... + abs(mrtt[N-1) - mmrtt) / N;

This is known as a "rolling average" and "rolling variance", and have 
been tried in a number of contexts before, in particular, for a ECG 
scanning system:

"An Experimental ECG Scanning System," (extended abstract)
K.P. Birman and J.D. Touch, Proc. Of the Ninth Annual Conf. Of the IEEE 
Engineering in Medicine and Biology, Boston, Nov. 13-16, 1987. The note 
on which this extended abstract is based is available at 
http://www.isi.edu/touch/pubs/tri.pdf

The use of 'calibrating factors' was used in that work to describe the 
difference between signal and noise - literally, signal was when the 
data exceeded the average by more than some multiple of the variance.

It has been raised as an alternative to TCP's capacitive decay model 
several times, notably at several IETF TCPIMPL meetings. Whether it is 
strictly more stable or useful than the current 'capacitative decay' 
model has not yet been proven.

Joe