Congestion Avoidance and Control

This paper is a summary of the congestion control mechanisms built into the TCP standard. One of these is slow start, where the packet rate is only increased when an ack is received. The other is additive increase/multiplicative decrease. Lastly, the exponential backup mechanism is briefly justified.

These algorithms assume a number of things. One is that a dropped packet indicates congestion. In highly intermittent networks (802.11) this doesn't hold, and these algorithms tend to grossly overestimate the congestion in the network.

The paper itself is great, concise and clear. It looks long, at 25 pages, though around half of it is appendix.

As far as discussion topics, my normal questions come out:
Competing systems, and why
Assumptions, and why

This leads to a discussion of DTN, UDP, and any other systems I don't know about.

I'm also curious about what other assumptions are implicit. The big one for me has been that packet dropping one, but various work have shown some security assumptions which do not necessarily hold.

This paper should totally be kept in the reading list.

This paper was mostly technical work, with little pure research. I don't feel as though it left many open research questions. If any, it would be attacks on this system and it's assumptions.