Efficient Placement of Serverless Applications across the Edge-Cloud Continuum

Abstract

With the advancements in real-time IoT applications, Function-as-a-Service (FaaS) platforms are becoming increasingly crucial as they offer auto-scaling capabilities, cost efficiency, and rapid deployment functionalities to end-users. Serverless applications can utilize both Edge and Cloud serverless platforms together for operations but it still has many challenges associated with it. Therefore, this thesis presents a comprehensive exploration of serverless computing across Edge-Cloud Continuum, beginning with a detailed performance analysis of the OpenWhisk platform deployed in Edge and Cloud environments to investigate critical factors affecting the performance of real-time serverless applications. This analysis then lays the foundation for the development of a novel framework, EPSA (Efficient Placement of Serverless Application), designed to optimize the deployment of serverless applications across the Edge-Cloud infrastructure. The primary objective is to minimize End-to-End (E2E) latency while satisfying other performance critical constraints including execution cost limit, location constraints, and application delay thresholds. EPSA incorporates a performance prediction module that utilizes both statistical and machine learning techniques to accurately and efficiently predict function performances, thereby facilitating more effective placement of applications. The approach includes a structured multi-level heuristic pipeline, where the first level identifies the pareto-front solutions, and the last level selects the best resource node for serverless function placement. Additionally, our framework integrates an adaptive placement strategy that dynamically adjusts the constraints based on real-time conditions to optimize the success rate of function placement. The combination of all these characteristics represents a unique and efficient approach for deploying serverless applications on FaaS Edge-Cloud infrastructure. The framework’s effectiveness is demonstrated using testbed setup that compares its performance against established benchmarks including OpenWhisk (baseline), Round Robin (RR), and Least Connection(LC). These evaluations show that our approach significantly reduces E2E latency and execution costs, increases placement success rates, and enhances Edge resource utilization, thereby decreasing total dependency on Cloud-based traditional infrastructures.