Function-Aware Resource Management Framework for Serverless Edge Computing

Abstract

Serverless edge computing is an emerging concept where only required functions are defined and executed as container instances at the edge cloud. The edge cloud has finite resources; therefore, sophisticated resource management is indispensable to accommodate more requests. In this article, we propose a function-aware resource management (FARM) framework for serverless edge computing that defines per-function queues to maximally utilize edge cloud resources. The FARM framework optimally determines: 1) which container instances should be maintained as warm status and 2) the amount of computing resources assigned to them. The FARM framework specifically formulates a constrained Markov decision process problem to minimize the memory resource consumption for the warm status maintenance while guaranteeing on-time task completion and converts it to a linear programming model to derive the optimal solution. The evaluation results show that the FARM framework can reduce the memory resource consumption of the edge cloud while meeting the on-time task completion.