Showing papers on "Complex network published in 1984"

Bounding algorithms for queueing network models of computer systems

Abstract: This thesis is concerned with the problem of efficiently analyzing large separable queueing network models of computer systems. Obtaining the exact solution of such a network is often infeasible. Approximation techniques have been proposed that significantly reduce the cost of network analysis; however, these techniques lack bounds on the errors that they introduce. This thesis proposes algorithms that bound, rather than only estimate, the queueing network solution. Each bounding algorithm is based on the iterative structure of one of the exact solution algorithms, in which a queueing network solution is iteratively computed from the solutions of simpler networks. In an exact solution algorithm, the iteration is started from trivial networks whose solutions are known a priori. By starting from appropriate approximate solutions of more complex networks, the proposed bounding algorithms significantly reduce the cost of network analysis. Each bounding algorithm is analyzed with respect to its computational cost and its accuracy. A comparison is given among proposed and previous bounding algorithms, with the goal of determining appropriate bounding algorithms for various accuracy requirements, cost constraints, and queueing network types.