Buffering and Flow Control in Message Switched Communication Networks.

摘要

The mathematical study of buffering and flow control is based on a gradual input queueing model. The gradual input model has been used previously to study data multiplexors. Here it is extended to an entire message switched communication network. A probability of buffer overflow analysis is developed and used to determine buffer requirements. A delay analysis is also developed. The results obtained using the gradual input queue are compared to the commonly used M/M/1 queue model for message switched networks. The gradual input model allows one to observe several effects due to a finite number of finite rate traffic sources in such networks that cannot be observed using the M/M/1 model. Flow control is studied in tree concentration structures. The flow control assures that buffer overflows will occur only at source nodes, not in the interior of the tree. The problem of finding the buffer allocation that minimizes the probability of buffer overflow in such a tree is studied. It is shown that in certain cases it is optimal to place all buffers at source nodes. This is, however, not always so and insight into this is given by example. Determining the performance of a tree structure in which flow control is being used is a difficult analytic problem. An approximate analysis based on a first passage time theorem for Markov chains is therefore developed for an example. The approximate analysis is verified by simulation. (Author)