EFFICIENT ALLOCATION OF DISTRIBUTED OBJECT-ORIENTED TASKS TO A PRE-DEFINED SCHEDULED SYSTEM

摘要

A key phase in the design process of software for distributed systems is the allocation of the software components to the available hardware. A problem arises when a software/hardware mismatch occurs. This paper presents a solution to that problem by introducing a technique that guarantees efficient allocation of predefined scheduled object-oriented software components to the available hardware based on genetic algorithm. The allocation is to be made dynamically in a system with a predefined schedule. We thus modified the parameters of the genetic search technique to allow converging to the best solution in a relatively short time to be suitable for the dynamicity of the allocation. The performance of the allocation technique is evaluated in terms of the time cost (CPU clock pulses for more generality) required for the GA search to converge to the optimal allocation structure of the software components. The results obtained by the proposed technique are compared against the results from the branch-and-bound search technique. The experimental results indicate the effectiveness of the proposed technique in reaching the optimal allocation in considerable time, showing that it requires much less time than branch-and-bound.