ServScale: Concurrency-Aware Serverless Execution and Scaling Paradigm

ServScale addresses the challenges of deploying applications with intra-instance concurrency on serverless platforms. Traditional serverless platforms often struggle with resource inefficiencies due to their reliance on horizontal scaling and static resource allocation. ServScale introduces a hybrid scaling mechanism that combines horizontal and vertical scaling to optimize resource utilization. The system consists of two main components: the Informer and the Coordinator. The Informer tracks the workload of functions and makes scaling suggestions based on real-time data. The Coordinator leverages this information to scale function containers both horizontally and vertically, ensuring precise resource allocation. ServScale's predictive prewarming mechanism minimizes scheduling latency and reduces the number of idle containers, further improving resource efficiency. The evaluation of ServScale demonstrates significant reductions in CPU and memory consumption compared to traditional serverless and IaaS-based systems. ServScale's hybrid elasticity mechanism makes it versatile for workloads with varying resource characteristics, making it a cost-effective solution for deploying applications with intra-instance concurrency.