PAPS: A Framework for Decentralized Self-management at the Edge

The emergence of latency-sensitive and data-intensive applications requires that computational resources be moved closer to users on computing nodes at the edge of the network (edge computing). Since these nodes have limited resources, the collaboration among them is critical for the robustness, performance, and scalability of the system. One must allocate and provision computational resources to the different components, and these components must be placed on the nodes by considering both network latency and resource availability. Since centralized solutions could be impracticable for large-scale systems, this paper presents PAPS (Partitioning, Allocation, Placement, and Scaling), a framework that tackles the complexity of edge infrastructures by means of decentralized self-management and serverless computing. First, the large-scale edge topology is dynamically partitioned into delay-aware communities. Community leaders then provide a reference allocation of resources and tackle the intricate placement of the containers that host serverless functions. Finally, control theory is used at the node level to scale resources timely and effectively. The assessment shows both the feasibility of the approach and its ability to tackle the placement and allocation problem for large-scale edge topologies with up to 100 serverless functions and intense and unpredictable workload variations.