Summary: Exploring Congestion Control 1
Aditya Akella 2 Srinivasan Seshan 3 Scott Shenker 4
Ion Stoica 5
May 2002
CMU­CS­02­139
School of Computer Science
Carnegie Mellon University
Pittsburgh, PA 15213
Abstract
From the early days of modern congestion control, ushered in by the development of TCP's and DECbit's
congestion control algorithm and by the pioneering theoretical analysis of Chiu and Jain, there has
been widespread agreement that linear additive­increase­multiplicative­decrease (AIMD) control algorithms
should be used. However, the early congestion control design decisions were made in a context where loss
recovery was fairly primitive (e.g. TCP Reno) and often timed­out when more than a few losses occurred
and routers were FIFO drop­tail. In subsequent years, there has been significant improvement in TCP's
loss recovery algorithms. For instance, TCP SACK can recover from many losses without timing out. In
addition, there have been many proposals for improved router queueing behavior. For example, RED active
queue management and Explicit Congestion Notification (ECN) can tolerate bursty flow behavior. Per­flow
packet scheduling (DRR and Fair Queueing) can provide explicit fairness.
In view of these developments, we seek to answer the following fundamental question in this paper: Does

Source: Akella, Aditya - Department of Computer Sciences, University of Wisconsin at Madison

Collections: Computer Technologies and Information Sciences