Supporting the Security Awareness of GA-based Grid Schedulers by Artificial Neural Networks

Abstract

Task scheduling and resource allocation remain still challenging problems in Computational Grids (CGs). Traditional computational models and resolution methods cannot effectively tackle the complex nature of Grid, where the resources and users belong to many administrative domains with their own access policies and users' privileges, and security and task abortion awareness are addressed as important scheduling criteria. In this paper we propose a neural network approach for supporting the security awareness of the genetic-based grid schedulers. Making a prior analysis of trust levels of the resources and security demand parameters of tasks, the neural network monitors the scheduling and task execution processes. The network learns patterns in input (tasks and machines initial characteristics) and outputs (information about resource failures and the resulting tasks and machines characteristics) data, and finally sub-optimal schedules are generated, which are then used to modify the initialization procedures of genetic scheduling algorithms. We extended the Hyper Sim-G Grid simulator framework by Neural Network module to evaluate the proposed model under the heterogeneity, the large-scale and dynamics conditions. The relative performance of GA-based and Neural Network GA-based schedulers is measured by the make span and flow time metrics. The obtained results showed the efficacy of the Neural Network approach to enhance the secure GA-based schedulers.