Energy-Aware Scheduling of a Serverless Workload in an ISA-Heterogeneous Cluster

The serverless model decouples computation from infrastructure. This results in flexibility for running serverless functions on heterogeneous hardware, such as emerging x86/ARM ISA-heterogeneous clusters. We present a method for scheduling serverless workloads across ISA-heterogeneity boundaries to reduce energy usage. Our method combines the offline profiling of functions for energy use and performance, the construction of performance/energy affinity models, and an energy-aware scheduler. Our evaluation with servers equipped with Xeon x86 and Ampere Altra Max ARM processors and 22 serverless functions shows that energy usage can be reduced by up to 15.2%.