Investigating Efficient Edge Offloading Architectures for Serverless Systems

Modern cloud-native applications use serverless computing, which enables cost- and resource-efficient scalable run-time behaviour. Adopting serverless computing in edge is becoming popular. Many of these emerging edge applications are time-critical, requiring quick responses. However, edge workloads often need to be dynamically offloaded to the cloud due to resource constraints. Consequently, identifying efficient offloading architectures is crucial to utilizing the benefits of serverless computing on the edge. In this work, we present and evaluate three offloading architectures under dynamic workload conditions for edge computing: (1) node-to-node, (2) cluster-to-cluster, and (3) centralized offloading. Furthermore, we highlight when offloading is an effective tool for increasing the efficiency of the edge. Our initial experiments show that in cases where client requests result in small computations, serverless computing on edge can improve response time by at least 36%.