[e2e] Timeout or dupacks?

[Peter Rabinovitch]

See the graph on page 14 of I.Kaj, J. Olsén, "Throughput modeling and
simulation for
single connection TCP-Tahoe", Full Paper, 17th International Teletraffic
Congress ITC'01, Salvador da Bahia, Brazil, September 24-28, 2001.
Available at http://www.math.uu.se/~ikaj/papers.html

It is a graph of (their model & simulation if I remember correctly)
overall probability of loss vs. probability that the loss was due to
timeout. As I recall, in conditions where there are less than 2% drops
for any reason, 90% of the drops are indicated by TD (triple duplicate
ACKs) and only 10% by TO (timeouts).


Tamas Elteto wrote:
> 
> Hi,
> 
>  Recent observations have found that, in many cases, retransmission due to
> timeout is a more frequent event - or at least as frequent - during a
> TCP transfer than retransmission due to dupacks. For example, a well known
> paper - "Modelling TCP Throughput: A Simple Model and its Empirical
> Validation" - shows measurements supporting this. However, I have
> different experience though I have not performed detailed studies yet.
> 
>  My question is the following: Does anybody know about studies that
> investigate the typical packet loss probabilities and proportion of
> timeouts and dupacks in the today's Internet?
> 
> Regards,
> Tamas