Performance Characterization of Containers in Edge Computing

2025-05-04

AI Summary

This paper addresses core challenges in containerized deployment on resource-constrained edge devices (e.g., Raspberry Pi), including high cold-start latency, severe I/O bottlenecks, frequent cache misses, and difficulty guaranteeing real-time performance. We conduct a systematic empirical study using a custom-built embedded physical testbed—equipped with environmental sensors and cameras—and evaluate Docker containers under both micro-benchmarks (CPU, memory, network) and macro-benchmarks (AI inference, sensor I/O). To our knowledge, this is the first work to quantitatively characterize container overhead and the isolation-efficiency trade-off in realistic edge settings. We derive lightweight, real-time-aware container configuration guidelines, identifying memory reservation, storage driver selection, and cgroups tuning as critical optimization levers. Results show that containerization incurs an average 210-ms cold-start delay, a 12% increase in page faults, and an 18% reduction in network throughput—establishing a reproducible empirical baseline and actionable deployment guidance for edge containerization.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Application Domains: Internet of Things, Sensor Networks & Smart Cities

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Web performance, measurement, and characterization
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Security and Privacy: Large-scale security measurements

Abstract

This paper presents an empirical evaluation of container-based virtualization on embedded operating systems commonly used in Internet of Things (IoT) deployments. Focusing on platforms like the Raspberry Pi, we investigate the feasibility and performance implications of deploying Docker containers in resource-constrained edge environments. Our study employs both microbenchmarks (CPU, memory, and network profiling) and macrobenchmarks (AI-driven inference, sensor IO workloads) to capture a comprehensive view of system behavior. The analysis is conducted on a custom-built physical testbed comprising Raspberry Pi devices equipped with environmental sensors and camera modules, enabling real-time deployment and measurement of representative IoT workloads. Through quantitative analysis across a diverse suite of IoT tasks and real-time application services, we identify key overheads introduced by containerization and characterize challenges specific to embedded IoT contexts, including limited hardware resources, cold-start delays, and suboptimal IO handling. Performance metrics include CPU utilization, memory faults, cache misses, network throughput, and latency. Our findings highlight trade-offs between isolation and efficiency and offer insights for optimizing container configurations to meet the real-time and reliability requirements of edge computing applications.

Problem

• Evaluating container performance on IoT embedded systems
• Assessing Docker feasibility in resource-limited edge environments
• Identifying containerization overheads in IoT hardware constraints

Innovation

• Evaluates Docker containers on Raspberry Pi IoT
• Uses microbenchmarks and macrobenchmarks for profiling
• Analyzes container overheads in edge environments