CSE 770 Paper Review

Reviewer: Amy Freestone
Date: 9-29-2005

How would you rate this paper, relative to others we have read? top 25%, but not top 10%

How would you rate your knowledge of the topic of this paper? novice

What problem or issue does the paper address? Why is it important?

The paper addresses the issue of congestion control with a small number of bits for feedback. It is important because congestion control is necessary for effective communication via the network and because achieving that control with a small enough number of bits means that existing header formats could be used without making them larger.

What are the main contributions of the paper and why are they important?

The main contribution of the paper is a method for congestion control which uses only two existing bits in the packet header for its feedback.

Other contributions include:

• VCP divides the network utilization into one of three regions, able to be encoded with two bits.
• The algorithm run by the controllers is determined by the network utilization region.
• The routers periodically estimate load factors for each outgoing link which are then used to determine which utilization region each link is part of. This data is then sent back to the sender via the receiver's ACK packets, producing no additional traffic.
• Unlike those using TCP, VCP flows' throughputs are not affected by the heterogeneity of their RTTs.
• VCP uses the scaling of its MI and AI parameters to help prevent oscillations between modes.

How significant are these contributions relative to previous work?

The existing TCP protocol uses congestion control via a method unsuited to networks with a high BDP.

The XCP protocol produces similar congestion control but in more bits, so it cannot be implemented without modifying the packet header size. VCP's congestion control operates at levels similar to XCP's but with less bits and can be implemented without modifying the packet header size. As such a modification would be quite difficult to make standard, this is quite a significant contribution.

VCP removes the effect of RTT heterogeneity that was found in TCP for not wildly dissimilar RTTs, making the control more balanced, a contribution of at least minor significance.

Give detailed comments justifying your view of the paper.

The authors do a good job explaining why they feel it necessary that a replacement for TCP be sought and why other control schemes are inadequate. They also extensively discuss what approaches of previous schemes work well and how they are building off of those ideas, as well as the key issues (determining the load factor transition point, setting the congestion control parameters, taking heterogeneous RTTs into account) that went into their design.

Improvements over previous schemes are clearly discussed, giving credibility to the author's claim that their scheme is generally better than its predecessors.

The authors provide varied situations under which they have tested VCP, giving a reader more confidence in the consistency of the performance of VCP.

I find somewhat troublesome the use of a model which the authors admit may not accurately reflect the dynamics of VCP.

The lack of an actual implementation of VCP makes it difficult to be certain that the theoretical results will be achieved in practice.