A Hybrid Fault Tolerance Model for Reliable Scheduling of Critical Real-Time Applications on Grid Systems
Bahman Arasteh, Amir Masood Rahmani, Saeed Taghavi Afshord · International Symposium on Parallel Architectures, Algorithms and Programming · 2010
Since the grid system is implemented on a network framework with heterogeneous remote resources, it is a hazardous environment. The possibility of failure occurrence on the grid resources are not negligible and should be considered. On the other hand, the reliability is a major criterion for the safety-critical and real time applications. This paper focuses on dependability aspects of grid scheduling strategies such as reliability and availability by using various forms of redundancy (e.g. hardware, software, and data). A hybrid method including fault masking and fault tolerance is used in this paper to improve the reliability of grid scheduling system. The proposed scheduling model integrates an active redundancy and backup majority voting together. After the detection of an error, multiple check-pointing as a recovery technique is invoked by the proposed scheduler. This method can be used at each level of scheduling based on a hierarchical and dynamic architecture. This scheduling model does not have a significant time overhead on other performance criteria such as waiting time, completion time and throughput. The proposed scheduling model also tries to reduce the resource cost in economic grids. Simulation results of proposed model in the presence of faults and resource failure show the improvement of reliability when compared with non-fault tolerance scheduler and other classic methods.