A features based expert artificial neural networks scheduling system

Abstract EN

Scheduling problems are one of the most challenging problems that can utilize the rapid improvements in parallel computer architectures.

In the context of scheduling, the problem is to exploit all the available resources of the available parallel architecture and optimize the target performance measure(s) or scheduling criteria.

The research work presented in this paper is motivated by the fact that there is no optimal scheduling approach due to the NP-complete nature of scheduling problems.

Moreover, the fact that scheduling of any problem on parallel / distributed platform is dependent on all the available inputs / features of that problem.

This paper introduces a scheduling model based on utilizing all available features that can be extracted from the problem then using them as inputs to a suite of expert artificial neural network architectures.

The proposed system was tested on a benchmark-scheduling problem in addition to a real time scheduling problem.

Regardless of its innovative set of features, and its unique architecture, the proposed system was compared with three other systems.

The experimental results showed that the proposed system improved the scheduling speedup of the given problems by 12% over the best of the other three systems.

In addition, the proposed system did not only enhanced the given problem performance measures but it did decrease the number of required processing elements and hence computational processing power for this real world problem.