Dynamic scaling improves latency and resource use in edge stream processing
A Dynamic Vertical Scaling Strategy for Distributed Stream Processing Applications in Edge Computing
Distributed, Parallel, and Cluster Computing
Summary
Distributed stream processing at the edge faces tough choices because devices have limited and different resources but need to process data quickly and smoothly. The authors developed a smart method using machine learning to decide how to best allocate tasks and resources dynamically, aiming to keep delays low most of the time. Their tests in simulations showed this method kept the data flowing well, violated latency goals very rarely, and often used less CPU compared to previous approaches. This helps edge computing systems work better under changing conditions.
What this means in practice
- •For edge system operators: Operate stream processing applications at the network edge more efficiently by dynamically adjusting resource use and task distribution to meet latency targets.
- •For cloud infrastructure engineers: Improve management of distributed stream processing workloads by applying machine-learned vertical scaling policies that adapt to changing resource availability and workload.
Authors
Guilherme Hiago Costa dos Santos, Carlos Henrique Kayser, Tiago Coelho Ferreto
Abstract
Distributed Stream Processing applications at the edge must reconcile low latency and high throughput with limited and heterogeneous resources. This paper presents a dynamic vertical scaling strategy based on Proximal Policy Optimization, formulated as a Partially Observable Markov Decision Process. The policy jointly adjusts task allocations and prioritizes compliance with a p95 end-to-end latency Service Level Objective. In EdgeStreamPy simulation experiments with two application profiles, two workloads, and ten paired placements per combination, the selected policies preserved throughput, obtained mean violation rates from 0.03% to 0.30%, below VRebalance in every scenario, and used less CPU in three of four combinations. The comparison covers complete controller configurations with different decision frequencies.