Solution of Markov Renewal Decision Processes with Application to Computer System Scheduling.
MICHIGAN UNIV ANN ARBOR SYSTEMS ENGINEERING LAB
Pagination or Media Count:
The paper pursues the theoretical development and application of Markov renewal decision processes in the design of efficient multiprogrammed computer systems. A Markov process is used to model the software design of multiprogrammed computer systems namely, the problem of assignment of central processors and main memory to those jobs requesting them. Application of the optimization method to this model allows the determination of scheduling rules which maximize throughput. Parameters which may be varied in the model include the size of main memory relative to the sizes of individual tasks. the execution characteristics of tasks themselves, the number of processors, and the fraction of its total memory requirement each task may share with other tasks. Results for a range of values of parameters are presented which compare the throughput possible under optimal scheduling rules under a good heuristic scheduling rule. Author
- Operations Research
- Computer Programming and Software
- Computer Hardware
- Computer Systems