Review: Congestion Control for HIgh Bandwidth-Delay Product Networks

Review: Congestion Control for HIgh Bandwidth-Delay Product Networks

Reviewer: Di Niu

This is a very solid paper which reconsiders the issue of congestion
control in modern networks and proposes XCP as a substitute for TCP.
The motivations for devising XCP are that TCP is not suitable for
high bandwidth-delay product networks and that TCP has inherent
deficiency in congestion control.

First, the insufficiency of TCP is pointed out. All congestion
control protocols in TCP become prone to instability in theory in
high bandwidth-delay product networks. TCP's additive increase policy
limits its ability to acquire spare bandwidth to one packet per RTT.
TCP is also insufficient concerning fairness.

Three objectives are proposed as the design rationale of congestion
control protocols. First, instead of using packet loss as a signal of
congestion, the network should be designed to avoid packet losses as
much as possible. Second, the aggressiveness of the sources should be
adjusted according to the delay in the feedback-loop. Third, the
dynamics of the aggregate traffic should be made independent from the
number of flows.

In the design of XCP, the routers inform the senders about the degree
of congestion explicitly. It decouples the utilization control from
fairness control. It does not need to maintain per-flow states.
Moreover, XCP only requires a few CPU cycles per packet, making it
practical even for high-speed routers. Control theory based analysis
supported the correctness of the efficiency controller and the
fairness controller. Extensive NS-2 based experiments have been
conducted to compare the performance of XCP with that of TCP.

In my view, this is a very exciting and solid paper which proposes
effective methods and provides solid support for these ideas. Many
useful ideas and tools can be used to benefit other researches. For
example, the idea of decoupling efficiency from fairness can be
applied to other areas such as load balancing and resource allocation
in P2P networks. The control theory based analysis could be an
effective tool to analyze linear systems with delays.

The only weakness of the paper seems to be the complexity of
deployment in current network environments.
Received on Mon Oct 02 2006 - 23:37:26 EDT