[p2pi] Thought on congestion

[Henning Schulzrinne <hgs at cs.columbia.edu>]

A quick thought on congestion: The discussion seemed to assume that we  
can throttle sources and make the problem go away. However, this  
depends on the network capacity. For the Columbia link, the network is  
essentially fully loaded all day, so the only shifting would be  
transferring load to the wee hours of the morning, which is presumably  
feasible only for Windows updates, large physics data sets, network  
backups and similar bulk date of no direct interest to humans. Even  
so, that would only yield an increase in overall throughput of maybe  
30%, i.e., barely a year's worth of typical traffic growth. In those  
circumstances, congestion control really means "you can't download  
that movie". It would be interesting to see the diurnal load curve for  
consumer networks.

Capping peak usage would still be quite useful, as would protecting  
real-time traffic, but I think we have to be more precise as to what  
congestion control really means and what amount of time shifting is  
necessary.

I was also a bit surprised by the notion that the top 5/10% of users  
should simply be cut off in some way. This is not economically  
efficient; they might, after all, be willing to pay for the benefit  
they get from all that bandwidth.

Henning
_______________________________________________
p2pi mailing list
p2pi at ietf.org
https://www.ietf.org/mailman/listinfo/p2pi