[e2e] Spurious Timeouts in mobile wireless networks, was: Re: Retransmission Timouts revisited

[Detlef Bosau]

I've learned some interesting thins on this matter in some offline
discussions.

In consequence, I know less then ever, but that's life....

Basically, I´ve learned that loss detection via RTO should only be the
last ressort.

Actually, loss detection should be done mostly by the observation of
ACKs, particularly SACK. Marc Allman refered rfcs 3517 and 3782.

In consequence, RTO estimators are not that important for the
performance of TCP. They must be large enough to avoid too many spurious
   timeouts. If they are a little bit too large, it's at least not
disastrous. They are only a last resort, if anything else fails.


The "spurious timeout" problem often claimed in mobile wireless networks 
thus should be basically a problem with extreme delay spikes.

As to my understanding, we can exclude other problems, the only problem 
with TCP in mobile wireless networks are unexpected, or unforeseeable 
respectively, delay spikes or a sudden increase of delays.

This can be due to
- scheduling (voice is priorized over data)
- route change (I don´t know whether this is that important, because 
much of this is micromobility)
- changes in physical path properties.




-- 
Detlef Bosau
Galileistrasse 30
70565 Stuttgart
Mail: detlef.bosau at web.de
Web: http://www.detlef-bosau.de
Mobile: +49 172 681 9937