> But it still looks that in some rare cases it could help. > For example, I heard that someone is actually evaluating benefits of > packet spacing over some kind of wireless networks: it seems that in > presence of large RTTs combined with network "blackouts", spreading > packets can improve performances.
Daniele,
Have you seen the FastTCP algorithm from CalTech (http://netlab.caltech.edu/pub/papers/FAST-infocom2004.pdf)? It consists of 2 independent parts. One is a new algorithm for messing with the congestion window (ie. Reno, Westwood, etc.), the other is a 'queue delay' algorithm. The queue delay algorithm (referred elsewhere in their article as 'Burstiness Control') they describe is rather similar to your patch (see p 5, bottom of first column for details).
Also, I suspect the impact of the algorithm extends beyond a few special cases. A doctoral candidate at Ohio State (a student of Wu Chi Feng) did his dissertation on the effects of burstiness and burstiness control on packet loss using network simulations (finished in '99 or '00). His conclusion was that burstiness control at the sender improved network throughput pretty much universally. If I remember correctly, his simulated network assumed that the primary reason for packet loss was lack of buffer space at an intermediate router. To the best of my knowledge, this finding has not been verified empirically. I apologize that I do not have the persons name and can't give a more solid reference for his dissertation.
Daniel McEnnis.
- : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
