Fast Congestion Notification Mechanism for Next Generation Routers

The aim of this thesis is to present a new proactive congestion control mechanism, namely "Fast Congestion Notification (FN)" for TCP/IP ECN-capable routers. FN has
been developed and implemented in Network Simulator 2 (ns-2). It uses the instan-taneous(current) queue length and the average packet arrival rate to make its control decisions. The new mechanism drops the arriving packets (if non-ECN) and marks packets (if ECN) at the head of the queue before the buffer overflows, to effectively control the current queue length (Qcur) below the required optimal queue length (Qopt) in order to reduce the queuing delay and avoid the buffer overflows; and to maintain the average packet arrival rate (R) about the outgoing transmission link capacity (U) in order to enable the congestion and queue length control. Currently, Random Early Detection
(RED) mechanism is used in the Internet. RED uses the average queue length for making the control decisions. The use of average queue length makes RED reacts to congestion slowly. This results in large queue length variation and untimely congestion detection and notification which would cause performance degradation due to high queuing delays and high packet loss. The combination of the instantaneous
(current) queue length and the average packet arrival rate used by FN showed superior performance to that of RED in term of fast congestion detection and notification.
The limitation of the new mechanism is that it works only with responsive connections which play a big role in avoiding and controlling the congestion. Since this thesis
considers the necessity for modern queue management mechanisms that can control the Internet traffic efficiently and improve the Internet performance, the major contribution of this thesis is to provide a new pro-active queue management mechanism that responds to congestion more quickly, delivers congestion notification timely, and
controls queue length directly to congestion which results in minimizing queue length variation. All these would help improve the Internet performance.