Fault-Tolerant Resource Provisioning with Deadline-Driven Optimization in Hybrid Clouds

Emmanuel Ahene, Kingsley Nketia, Heyang Xu
2016 International Journal of Advanced Computer Science and Applications  
Resource provisioning remains as one of the challenging research problems in cloud computing, more importantly when considered together with service reliability. Fault-tolerance techniques such as fault-recovery is one of the techniques that can be employed to improve on service reliability. Technically, fault-recovery has obvious impact on service performance. Such impact requires detailed studies. Only few works on hybrid cloud resource provisioning address fault recovery and its impact. In
more » ... nd its impact. In this paper, we investigate the problem of resource provisioning in hybrid Clouds, considering the probability of hybrid cloud resource failure during job execution. We formulate this problem as an optimization model with operational cost as an objective function subject to the deadline constraint of jobs. Based on our proposed optimization model, we design a heuristic-based algorithm called dynamic resource provisioning algorithm (DRPA). We then perform extensive experiments to evaluate performance of the proposed algorithm based on a real world set of data. The results confirm the obvious impact of fault recovery on the performance metrics (operational cost and deadline violation rate) and also confirms that DRPA can be useful in minimizing operational cost. We present a hybrid Cloud model that enables the scalability of the resource-base of a software-as-a-service SaaS provider who intends to leverage resources in both private and public Cloud for job execution. We propose DRPA to address the problem of operational cost minimization associated with the leveraging of hybrid Cloud resources, considering the probability that resources (i.e.Virtual machines (VM) and communication links) may fail and recover. In addition, we take into account some practical issues such as communication cost, cost incurred at local (private cloud) and jobs that are dynamic in nature. Finally, we perform extensive experiments to evaluate the performance of the proposed algorithm in terms of operational cost and deadline violation rate.
doi:10.14569/ijacsa.2016.071250 fatcat:7z4tfonmwveu7ea7wns4jeclyi