Link Capacity Control in a Computer Communication Network.
Abstract
This paper describes a line switching communication network where, as the network becomes congested, additional lines are opened to relieve congestion. It is assumed that a fixed charge is paid for each line opening or closing operation; a cost is paid per unit time for each line in use; a cost is paid per unit time for packet storage in a packet queue; and reward is earned for each packet transmission completed. Under assumptions concerning Poisson arrival and service, a Markov model can be formulated which determines the average earning rate for the network under a specific policy. The policy is the control algorithm which specifies when lines should be opened or closed. The solution technique involves replacing an infinite subset of the state space by a finite set of states with equivalent behavior. A traditional technique called policy iteration can then be applied to the reduced finite model. The algorithm solves for th optimal policy, i.e., the policy yielding the highest earning rated. The paper illustrates how optimal policies for the finite optimization problem are also exactly correct for the original infinite problem. While the work describes how alone switching network can be modelled and optimal control strategies determined; it also illustrates a modelling technique which can be applied to other optimization problems having an infinite state space.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 1980
- Accession Number
- ADA123942
Entities
People
- Michael Schlansker
- Timothy Chou
Organizations
- University of Illinois Urbana–Champaign