Run-time application migration using checkpoint/restore in userspace
Aleksandar Tošić
TL;DR
This paper empirically evaluates the feasibility of using Checkpoint/Restore In Userspace (CRIU) to migrate running applications between hosts in edge and cloud environments, focusing on Docker-enabled deployments. It systematically measures how the memory footprint of a container affects checkpoint and restore times and monitors host CPU/RAM usage through system telemetry. The key findings are that checkpoint time scales linearly with the container's memory, restore time is less sensitive to memory, and CPU utilization does not scale with memory size, highlighting memory footprint as the primary performance driver. The work suggests CRIU is well-suited for stateless microservices on edge devices, while indicating limitations for high-availability servers requiring continuous uptime, and points to future work on memory optimization and orchestration integration.
Abstract
This paper presents an empirical study on the feasibility of using Checkpoint/Restore In Userspace (CRIU) for run-time application migration between hosts, with a particular focus on edge computing and cloud infrastructures. The paper provides experimental support for CRIU in Docker and offers insights into the impact of application memory usage on checkpoint size, time, and resources. Through a series of tests, we find that the time to checkpoint is linearly proportional to the size of the memory allocation of the container, while the restore is less so. Our findings contribute to the understanding of CRIU's performance and its potential use in edge computing scenarios. To obtain accurate and meaningful findings, we monitored system telemetry while using CRIU to observe its impact on the host machine's CPU and RAM. Although our results may not be groundbreaking, they offer a good overview and a technical report on the feasibility of using CRIU on edge devices. This study's findings and experimental support for CRIU in Docker could serve as a useful reference for future research on performance optimization and application migration using CRIU.