Review: DCCP

Paper: Designing DCCP: Congestion Control Without Reliability

Name: Robert Danek.
Course: CS2209, Fall '06.

    This paper discusses the design of the Datagram Congestion Control
Protocol (DCCP), a UDP-like protocol with congestion control but no
reliability.

    The motivation for designing the protocol arises from the
observation that applications, such as streaming media, are using UDP
more and more for long-lived flows. These applications do not require
reliability, since they would prefer sending a new packet over
retransmitting an old one. However, along with this comes lack of any
congestion control, which is bad for the network as a whole.

    The paper provides a description of the design of DCCP and the
problems encountered while designing it. The main goals of the
functionality provided by DCCP were as follows: it was to be a
minimalist design, so that it wouldn't be over engineered by providing
functionality that could and should be layered above it; it had to be
robust, so that denial-of-service attacks, and other similar attacks by
hackers, wouldn't completely break the protocol; and it had to be able
to meet the needs of a diverse range of applications, and as such should
only provide a framework for congestion control, as opposed to one fixed
algorithm.

    DCCP is a connection-oriented protocol and has acknowledgments
flowing between receiver and sender. The reason for this is because
congestion control requires loss detection, which cannot be done without
notifying the sender that packets arrived at the receiver.

    The authors go on to describe how they split the connection
management features from the congestion control mechanism. Connection
management involves such items as feature negotiation so that the
congestion control mechanism to be used can be decided upon. The
congestion control mechanisms that an application using DCCP has a
choice of includes CCID-2, which is a TCP-like congestion control
mechanism that uses congestion windows; and CCID-3, a TFRC congestion
control mechanism that uses a sending rate instead of a congestion window.

    This was a good paper. Along with delivering an informative
description of the new protocol being proposed, the authors also
provided insight into the tight-coupling of TCP's reliability mechanism
with its congestion control mechanism.
Received on Tue Oct 10 2006 - 00:04:29 EDT