Task Migration in Volunteer Computing Systems

Abstract

Volunteer Computing is the use of underutilised computing power donated by volunteers who want to participate in ongoing high-throughput scientific projects. Due to the general simplicity of this concept, this type of computing attracts thousands of volunteers from all over the world. Many of them, however, leave the network, prompting the main server to initiate Task Migration, which involves transferring the remaining job to another volunteer’s device on the network. In this paper, we conduct two experiments in the task migration procedure in peer-to-peer volunteer network. In the first experiment, the server starts the migration procedure when it detects the first available volunteer, whereas in the second one, the server starts the migration after all of the server’s sub-peers have completed their jobs. To ensure the network’s heterogeneity, we select the dataset GWA-T-13 Materna, which contains performance metrics described as trace files of over 1500 VMs from the distributed Materna Data Centers in Dortmund, Germany. In both experiments, we compare the total execution time of the entire task and the maximum execution time that a peer can have to complete its assigned work. The simulation results show that in the second experiment, the main task was completed in less time than in the first experiment.