A Reliability Aware Algorithm for Workflow Scheduling on Cloud Spot Instances Using Artificial Neural Network
Hoda Ghavamipoor, Sayyed Ali Kianian Mousavi, Hamid Reza Faragardi, Nayereh Rasouli · 2020
IT infrastructures are rapidly growing due to the increased demand for computing power used by applications. Furthermore, modern cloud data centers are hosting various advanced applications based on the user's needs. The goal of this paper is to maximize the reliability of running workflow applications considering spot instances revocation without imposing fault-tolerance overhead. For this purpose, we use an Artificial Neural Network algorithm (ANN) to define a failure prediction module for the cloud spot instances. Indeed, we introduce a novel workflow scheduling algorithm, named Reliability Aware and modified HEFT (Heterogeneous Earliest Finish Time) for minimizing the makespan of a given workflow subject to a specified reliability of the application. To evaluate our ANN based prediction model, we have used a benchmarking data set. The results of the model training demonstrate that the prediction accuracy is about 96 percent. Our experiments illustrate that although makespan does not improve significantly in all experiments, an acceptable level of reliability is achieved.