For Full-Text PDF, please login, if you are a member of IEICE,|
or go to Pay Per View on menu list, if you are a nonmember of IEICE.
Performance Analysis of Job Scheduling Policy for Interference Avoidance Using Stochastic Petri Nets
IEICE TRANSACTIONS on Fundamentals of Electronics, Communications and Computer Sciences
Publication Date: 1991/10/25
Print ISSN: 0916-8508
Type of Manuscript: Special Section PAPER (Special Issue on Petri Nets and Discrete Event Systems)
Full Text: PDF>>
In many parallel processing systems, interference among jobs, which comes from resource contention, causes performance degradation. In order to solve the problem of resource contention, an important question is: "How shall we optimize jop scheduling policy?" However, the traditional queuing theory assuming a first-come-first-served policy is not suitable for analyzing scheduling policy. A new method called Stochastic Petri Nets (SPN) is suitable for analyzing scheduling policy, because with it job arrival states, resource assignment states, and job flow due to these states can be modeled. However, the describing powers of places and transitions, elements of SPN, are as weak as Assembler in program languages. If we model scheduling logic in detail with SPN, that model becomes huge with a great number of places and transitions. Not only modeling of scheduling logic itself becomes difficult, but also analyzing of the model becomes difficult. In this paper, we consider a system in which resource contention causes job interference, and propose a method of analyzing scheduling policy for interference avoidance. In this method, for each scheduling policy, we introduce functions between job execution performance and system states. By incorporating this function into a SPN model, we cut down the size of the model. We apply the proposed method to analyzing performance of the control policy of a magnetic tape library, and show the efficiency and limit of this method.