DESA: Decentralized Self-adaptive Horizontal Autoscaling for Bursts of Load in Fog Computing

With the increase of Web of Things devices, fog computing has emerged as a promising solution to lower the communication overhead and congestion in the cloud. In fog computing systems, microservices are deployed as containers, which usually require an orchestration tool like Kubernetes for service discovery, placement, and recovery. One key challenge in the orchestration of microservices is establishing service elasticity in case of unpredictable bursts of load. Commonly, a centralized autoscaler in the cloud dynamically adjusts the number of microservice instances depending on the metric values monitored from distributed fog nodes. However, monitoring an increasing number of microservice instances can cause excessive network overhead and delay the scaling reaction. We propose DESA, a DEcentralized Self-adaptive Autoscaler through which each microservice instance makes its own scaling decision adaptively, cloning or terminating itself through a self-monitoring process. We evaluate DESA in a simulated fog computing environment with different numbers of fog nodes. The results show that DESA successfully reduces the scaling reaction time in large-scale fog computing systems compared to the centralized approach while resulting in a similar maximum number of instances and average CPU utilization during a burst of load.