Downtime Optimized Live Migration of Industrial Real-Time Control Services

TL;DR

This paper presents a novel live migration approach tailored for industrial real-time control services, significantly reducing downtime to sub-millisecond levels while maintaining reasonable migration times.

Contribution

It introduces a new migration concept that combines existing methods and virtualization to optimize for minimal service downtime in industrial settings.

Findings

Achieves sub-millisecond service downtime.

Migration time ranges from hundreds of milliseconds to two seconds.

Effective for high-demand real-time control applications.

Abstract

Live migration of services is a prerequisite for various use cases that must be fulfilled for the realization of Industry 4.0. In addition, many different types of services need to provide mobility and consequently need to be migrated live. These can be offloaded algorithms from mobile devices, such as unmanned vehicles or robots, security services, communication services or classic control tasks. In particular, the latter place very high demands on determinism and latency. Here, it is of utmost importance that the downtime of the service is as low as possible. Since existing live migration approaches try to optimize multiple metrics such as downtime, migration time as well as energy consumption, which are equally relevant in the IT domain, it is not possible to use any of these approaches without adoptions. Therefore, a novel concept is proposed that builds on top of both existing migration approaches as well as virtualization technologies and aims primarily at minimizing service downtime. Furthermore, the concept is evaluated using a test environment. The results show that a sub-millisecond downtime can be achieved with the proposed concept. Moreover, the total migration time is in the range of several hundred milliseconds for the highest performance setting and two seconds for a non-invasive approach.