A heuristic procedure for allocating tasks in fault-tolerant distributed computer systems

One way of achieving the increased levels of system reliability and availability demanded by critical computer-based control systems is through the use of fault-tolerant distributed computer systems. This article addresses the problem of allocating a set of m tasks among a set of n processors in a manner that will satisfy various task assignment, system capacity, and task scheduling constraints while balancing the workload across processors. We discuss problem background, problem formulation, and a known heuristic procedure for the problem. A new solution-improving heuristic procedure is introduced, and computational experience with the heuristics is presented. With only a modest increase in the amount of computational effort, the new procedure is demonstrated to improve dramatically solution quality as well as obtain near-optimal solutions to the test problems.     

A new approach for dealing with the startup problem in discrete event simulation

This article proposes a practical, data-based statistical procedure which can be used to reduce or remove bias owing to artificial startup conditions in simulations aimed at estimating steady-state means. We discuss results of experiments designed to choose good parameter values for this procedure, and present results of extensive testing of the procedure on a variety of stochastic models for which partial analytical results are available. The article closes with two illustrations of the application of the procedure to more complex statistical problems which are more representative of the kinds of purposes for which real-world steady-state simulation studies might be undertaken.     

Approximate Bayesian intervals for the reliability of series systems from mixed subsystem test data

The determination of exact Bayesian intervals for the reliability of a series system from subsystem test data gives rise to computational difficulties involving severe loss of computing precision as the number of subsystems in the system increases. The end points of Bayesian intervals are percentage points of the posterior distribution and these are shown to be well approximated by Cornish and Fisher expansions when the number of subsystems is small. As the number of subsystems in the system increases even greater accuracy is guaranteed by the asymptotic nature of the expansions. The system posterior distribution function is also shown to be well approximated by an Edgeworth expansion.     

Book Reviews

Strategic Views from the Second Tien The Nuclear Weapons Policies of France, Britain and China. Edited by J.C. Hopkins and W Hu, New Brunswick and London, Transaction Publishers, (1995) ISBN 1-56000-7907. $ 21.95

Multinational Military Forces: Problems and Prospects. By Roger H. Palin. Adelphi Paper 294, 11SS/Oxford University Press (1995) ISBN 0-19-828025-4 ISSN 0567 932X

South Africa in the Global Economy. Edited by Greg Mills, et al. South African Institute of International Affairs, Johannesburg, (1995) ISBN 1-874890-58-7. R. 45.60

The Limits of Air Powen The American Bombing of North Vietnam. By Mark Clodfelter, The Free Press. New York (1989). ISBN 0-02-905990-9. $22.95.

Guy Gibson. By Richard Morris with Colin Dobinson. Viking. London (1994). ISBN 0-670-82878-5. £18.00.

Dambuster: A Life of Guy Gibson VC. By Susan Ottway. Leo Cooper, London (1994). ISBN 0-85052-427. £16-95.     

Control-variate selection criteria

For multiresponse simulations requiring point and confidence-region estimators of the mean response, we propose control-variate selection criteria that minimize mean-square confidence-region volume in two situations: (a) Only the mean control vector is known, and standard linear control-variate estimation procedures are used. (b) Covariances among controls are also known and are incorporated into new linear control-variate estimation procedures. An example illustrates the performance of these selection criteria.     

Sandwich approximation of univariate convex functions with an application to separable convex programming

In this article an algorithm for computing upper and lower ? approximations of a (implicitly or explicitly) given convex function h defined on an interval of length T is developed. The approximations can be obtained under weak assumptions on h (in particular, no differentiability), and the error decreases quadratically with the number of iterations. To reach an absolute accuracy of ? the number of iterations is bounded by

     

All-integer linear programming — a new approach via dynamic programming

An exact method for solving all-integer linear-programming problems is presented. Dynamic-programming methodology is used to search efficiently candidate hyperplanes for the optimal feasible integer solution. The explosive storage requirements for high-dimensional dynamic programming are avoided by the development of an analytic representation of the optimal allocation at each stage. Computational results for problems of small to moderate size are also presented.