A Binary Feedback Scheme for Congestion Avoidance in Computer Networks with Connectionless Network Layer

We propose a scheme for congestion avoidance in networks using a connectionless protocol at the network layer. The scheme uses feedback from the network to the users of the network. The interesting challenge for the scheme is to use a minimal amount of feedback (one bit in each packet) from the network to adjust the amount of traffic allowed into the network. The servers in the network detect congestion and set a congestion indication bit on packets flowing in the forward direction. The congestion indication is communicated back to the users through the transport level acknowledgement. The scheme is distributed, adapts to the dynamic state of the network, converges to the optimal operating point, is quite simple to implement, and has low overhead while operational. The scheme also addresses a very important aspect of fairness in the service provided to the various sources utilizing the network. The scheme attempts to maintain fairness in service provided to multiple sources. This paper presents the scheme and the analysis that went into the choice of the various decision mechanisms. We also address the performance of the scheme under transient changes in the net1 I n t r o d u c t i o n Congestion in computer networks is a significant problem due to the growth of networks and increased link speeds. Flow and congestion control are problems that have been addressed by several researchers in the past [GK80]. With the increasing range of speeds of links and the wider use of networks for distributed computing, effective control of the network load is becoming more important. The lack of control may result in congestion loss, and with retransmissions, may ultimately lead to congestion collapse [Kle78]. The control mechanisms adopted to control the traffic on computer networks may be categorized into two distinct types: flow control and congestion control. End-to-end flow control mechanisms are used to ensure that the logical link has sufficient buffers at the destination. It is thus a "selfish" control function. Control mechanisms for congestion, on the other hand, address the "social" problem of having the various logical links in the network cooperating to avoid congestion of the intermediate nodes that they share. This paper proposes a mechanism for effective control in connectionless networks. We distinguish between congestion control, which has been studied in the past [BG85], [Nag84], [Jai86], and congestion avoidance. Congestion avoidance operates the network at the knee of the response time curve. This is the point at which the increase in throughput is small, ACM SIGCOMM -139Computer Communication Review while the response time increases rapidly with load. This enables the network to significantly reduce the probability of packet loss and preventing the possibility of serious congestion developing and impacting user performance in the network. A more detailed discussion of the differences is made in [JR88]. The congestion avoidance policy we propose here drives the operation of the network toward the knee of the delay curve. To achieve this operating point, the network provides some type of feedback so that the users may control the amount of traffic they place on the network. Congestion control mechanisms have been proposed that detect whether the network has gone beyond the cliff[Jai86], [BG85]. The feedback indicating congestion in the network is the loss of packets and the resulting time-out while waiting for the acknowledgment. Other forms of feedback of congestion information have also been used. An example is to send 'choke' or 'source quench' packets to control congestion [Nag84],