Reactive vs Predictive Live Migration in Edge Cloud

Migrating services in an edge-cloud environment poses unique challenges, including heterogeneous environments, potential failures, and uneven resource distribution. This paper studies and evaluates reactive and predictive migration approaches to support live migration in case of edge cloud computing failures. Telemetry information related to edge cloud computing has been considered to trigger migration, whereas deadlock prevention algorithm has been used to determine and select the target device to migrate services. The paper evaluates these strategies by comparing resource utilization, assessing differences between predictive and reactive migration and handling multiple migrations for tenants hosting several applications. Experimental results have shown that predictive migration can reduce the downtime of the hosted services. Additionally, the total migration cost can be increased for both scenarios where the containers can be migrated to different edge devices due to lack of available resources.