Optimization Analytic performance models are very well suited as kernels in optimization problems. Two major categories of optimization problems are static and dynamic optimization. In the former, performance measures are computed separately from an analytic queueing or CTMC model and treated simply as functions (generally complex and non-linear) of the control (decision) variables. In the latter class of problems, decision variables are integrated with the analytic performance model and hence optimization is intimately connected with performance evaluation. We limit our discussion to static optimization. . | Queueing Networks and Markov Chains Gunter Botch Stefan Greiner Hermann de Meer Kishor S. Trivedi Copyright 1998 John Wiley Sons Inc. Print ISBN 0-471-19366-6 Online ISBN 0-471-20058-1 1 1 JI Jr Optimization Analytic performance models are very well suited as kernels in optimization problems. Two major categories of optimization problems are static and dynamic optimization. In the former performance measures are computed separately from an analytic queueing or CTMC model and treated simply as functions generally complex and non-linear of the control decision variables. In the latter class of problems decision variables are integrated with the analytic performance model and hence optimization is intimately connected with performance evaluation. We limit our discussion to static optimization. For a treatment of dynamic optimization in the context of computer and communication systems see PGTP96 DrLa77 MeDü97 MTD94 MeFi97 MeSe97 and Meer92 . Designs of early computer and communication systems made little use of mathematical optimization techniques. Because the systems were not so complex the number of possible design alternatives was limited and a good design decision was often obvious. This state of affairs has changed dramatically since modern systems are very complex and requirements for high performance and dependability have evolved. Therefore an optimal design is not obvious any more and the number of possible design decisions can be enormous. A system designer is called on to produce a design of an implementable system from a given design specification. The design specification may include cost performance and dependability specifications in addition to functional specifications. The performance specification states how well the specified functions as stated in the functional specification should perform in terms of throughput response time etc. The dependability specification states how the system tolerates component sub-system faults. The dependability spec-
