Review:DCCP

Review: DCCP

Reviewer: Di Niu

This paper proposes DCCP, a congestion-controlled unreliable
transport protocol. Based on UDP, DCCP borrows ideas from congestion
control in TCP and makes them suitable for transmissions without
reliability. DCCP is primarily motivated by the emerging applications
at that time, such as media streaming and internet telephony. In
these applications, timeliness is preferred over reliability. The
paper is quite interesting, in the sense that it sounds original and
is the first to incorporate congestion control in UDP.

Five goals are first presented in the design of DCCP, including
minimalism, robustness, an adaptive framework for congestion control,
self-sufficiency and the support of time-reliability tradeoffs.
Nonetheless, DCCP deliberately omits multicast in its design goals.
This could be a weakness, as many streaming applications will benefit
from multicast.

In the presentation of design details of DCCP, the paper always
compares DCCP with TCP to show why specific modification is made to
TCP congestion control. In DCCP, sequence numbers refer to packets
instead of bytes. And the ACK-Vectors are used to acknowledge packets
received. Also, in DCCP only the most recently received packets have
ACKs sent in order to avoid the infinite regression of ACKs. In
addition, DCCP supports some explicit synchronization mechanism.
Sequence number length and DCCP's robustness against attacks are also
discussed.

However, one possible weakness of the paper is that the experimental
study is not strong enough. While the title sounds great at the first
glance, proposing congestion control in UDP, yet the content is no
more than a meticulous modification to TCP congestion control
protocols. And no extensive experiments are carried out to evaluate
such a modification.
Received on Mon Oct 09 2006 - 23:28:27 EDT