Performance Evaluation of the Computer Network Dynamic Congestion Table Algorthm

• Authors: C. A. Niznik
• Venue: IEEE Transactions on Computers
• Year: 1984

A comparison of the performance of the Dynamic Congestion Table (DCT) Algorithm, a computer network congestion control algorithm's delay table entry generation subalgorithm, is made here with a Pascal implementation of the present ARPANET SPF Algorithm (infinite buffer). The parameters of global throughput, delay, and power versus load are evaluated for 10 and 20 node sections of the ARPANET, using M/M/1 service and imterarrival distributions for the computer node (IMP) buffer to obtain computer node mean waiting time. The communication processor and channel delays are determined by the respective probabilistically weighted computer node mean waiting time. Point to point message path delays in the computer network are modeled by the repetitive occurrence of these three component delays in each path. The mathematical structure of two Moore Probabilistic Automatons (MPA) represents the three basic computer network component delays in each path. The probability values of the entries of each of the component MPA are converted to delay entries by the queueing theory node buffer delay computation. The end to end path delay is obtained from a cascaded MPA probability transition matrix, i.e., the multiplication of the individual computer network component MPA matrices. The DCT Algorithm is topology independent and therefore applicable to all centralized and distributed store and forward computer networks. The results indicate higher global throughput and lower global delay for a given load, and significant storage reduction using the DCT Algorithm computation compared to the current ARPANET SPF routing algorithm delay measurement technique.