Enabling High Availability Edge Computing Platform

TL;DR: This paper proposes to design a distributed edge computing platform by providing an easy management solution for remote distributed edge nodes and proposes a real-time internal and external container migration scheme to achieve cooperative processing, load balancing, data backup, and service failover.

Abstract: The prodigious data generated by all kinds of smart devices creates the need for an efficient and reliable platform which should be capable of managing huge data traffic. To overcome the acute load on the cloud and achieve low latency for real-time applications, the computation has been moved to distributed edge devices, i.e., fog/edge computing. However, the capability of end devices is not on par with the cloud. To address performance and reliability issues of these edge nodes, it is important to have one management platform that can solve reliability challenges and improve the overall service availability. In this paper, we propose to design a distributed edge computing platform by providing an easy management solution for remote distributed edge nodes. The major functions of our platform include service auto discover, lightweight container provisioning, GPU container support for deep learning applications, manage all active end nodes without a fixed IP address, secure remote deployment of edge analytic applications, and joint edge/cloud data communication and processing. To minimize the service downtime and guarantee high availability of edge services, we propose a real-time internal and external container migration scheme to achieve cooperative processing, load balancing, data backup, and service failover. Comparing with existing edge platforms, our solution achieves lightweight service management, low overhead, flexible and secure networking, and the significant reduction in latency.