Task scheduling and workload orchestration in edge computing

Abstract

An Edge Computing (EC) paradigm improves the quality of user experience when dealing with low latency and real-time interactive applications. The technology allows cloud services to be brought to the edge of the network closer to users rather than remote cloud platforms. Some key advantages provided by EC include reduced transmission delays, stable network connectivity and low latency. However, EC suffers from several challenges including under or over utilization of edge servers, task failures due to user mobility and network failures when connecting to the cloud. There is need to arduously harness some key resources of EC including virtual machine (VM) and wide area network (WAN) bandwidth in order to minimize those challenges. This can be achieved by task scheduling and workload orchestration mechanisms that efficiently exploit these resources. In this research, a task scheduling and workload orchestration problem was formulated and solved by genetic based and resource-aware algorithms. Particularly, a task migration scheme and a genetic algorithm (GA) to lower task failures due to user mobility and improve edge server VM utilization were developed. Furthermore, a resource-aware workload orchestration (RAWO) approach that considered VM utilization and WAN link bandwidth for task scheduling and offloading to edge servers and the cloud was modeled. Simulation results revealed that incorporating a task migration scheme and the proposed genetic algorithm scheduling approach minimized task failures and improved the overall edge server VM utilization. Results also showed that the RAWO approach reduced task failures due to VM utilization capacity and WAN congestion. This was due to the fact that the approach considered the average VM utilization and the WAN bandwidth while offloading tasks to edge servers and the cloud servers. Comparisons between the singletier and two-tier EC architectures were also carried out. Results showed that higher VM utilization and lower task failures were achieved in the two-tier EC as compared to single-tier EC for a high number of users. This was because the architecture offloaded tasks to both edge servers and cloud servers.