A package-aware scheduling strategy for edge serverless functions based on multi-stage optimization

Serverless computing offers a promising deployment model for edge IoT applications. However, server -less functions that rely on large libraries suffer from severe library loading latency when containerized, which is unfriendly to edge latency-sensitive applications. Most function offload strategies in edge environments ignore the impact of this latency. We also found that the measures taken by serverless platforms to reduce loading latency may not work in edge environments. To remedy that, this paper proposes a function offloading strategy to minimize loading latency, a new way to deeply integrate placement optimization with cache optimization. In this way, we first design a package caching policy suitable for edge environments based on the consistency of execution topology. Then a Double Layers Dynamic Programming algorithm (DLDP) is proposed to solve the problem of function offloading considering the dependent packages using a multi-stage progressive optimization approach. The caching policy is embedded in the scheduling algorithm through a phased optimization approach to achieve joint optimization. Extensive experiments on the cluster trace from Alibaba show that DLDP reduces the loading latency of packages by more than 97.84% and significantly outperforms four baselines in the application completion time by more than 55.67%.(c) 2023 Elsevier B.V. All rights reserved.