Congestion Management with feedback queue
A. Abi Khaled · 2011
The phenomenon of HOL (Head of Line) blocking occurs in communication where the data packets are under some conditions stopped waiting for the appropriate time for routing. This will cause communication delay in the network. Considerable research has been initiated by the demand for low-cost network topologies with no internal blocking (non-blocking topologies), capable to switch in parallel any possible combination of one-to-one, input and output connections. In order to be able to dynamically share the resources of the network, packet switched networks depend on statistical multiplexing. When several packets demand access to the same link, some of them maybe delayed waiting for other packets to move first. The short term or longer term delay causes degradation in the throughput of a certain network. When a link, route or destination is overloaded, while the demand on it is still increasing, congestion appears and latency increases. This will simply increase the demand of the link, route or destination beyond its capacity and show degradation in throughput and performance. This thesis investigates the handling of a classical and hard problem congestion management, reaching the suggestion of an alternative solution. In this work, a solution for congestion caused by head of line HOL blocking is proposed. An extensive taxonomy on available solutions is offered to the reader. We are able to present a solution with a feedback queue added to the input queue on a given switch in the network. A series of multiplexers control the routing of packets within the input, output, the main and feedback queue. We present an algorithm that controls the function of the multiplexers in order to eliminate the Head of Line blocking. The proposed algorithm proves to be a promising method as with the low cost of one additional queue, results close to the best case scenario have been achieved, performance improved and packet latency lowered to a margin close to the best case scenario.