Workload management for dynamic mobile device clusters in edge femtoclouds

Abstract

Edge computing offers an alternative to centralized, in-the-cloud compute services. Among the potential advantages of edge computing are lower latency that improves responsiveness, reduced wide-area network congestion, and possibly greater privacy by keeping data more local. In our previous work on Femtoclouds, we proposed taking advantage of clusters of devices that tend to be co-located in places such as public transit, classrooms or coffee shops. These clusters can perform computations for jobs generated from within or outside of the cluster. In this paper, we address the full requirements of workload management in Femtoclouds. These functions enable a Femtocloud to provide a service to job initiators that is similar to that provided by a centralized cloud service. We develop a system architecture that relies on the cloud to efficiently control and manage a Femtocloud. Within this architecture, we develop adaptive workload management mechanisms and algorithms to manage resources and effectively mask churn. We implement a prototype of our Femtocloud system on Android devices and utilize it to evaluate the overall system performance. We use simulation to isolate and study the impact of our workload management mechanisms and test the system at scale. Our prototype and simulation results demonstrate the efficiency of the Femtocloud workload management mechanisms especially in situations with potentially high churn. For instance, our mechanisms can reduce the average job completion time by up to 26% compared to similar mechanisms used in traditional cloud computing systems when used in situations that suggest high churn.