Towards a More Reliable and Available Docker-based Container Cloud

TL;DR

This paper introduces HYDRA, a design that isolates Docker containers from the daemon to improve reliability and availability in container clouds, addressing current limitations in Docker's architecture.

Contribution

HYDRA's novel design isolates containers from the Docker daemon, significantly enhancing container uptime during daemon failures or upgrades.

Findings

HYDRA achieves higher container availability.

HYDRA imposes moderate overheads.

HYDRA effectively isolates containers from daemon failures.

Abstract

Operating System-level virtualization technology, or containers as they are commonly known, represents the next generation of light-weight virtualization, and is primarily represented by Docker. However, Docker's current design does not complement the SLAs from Docker-based container cloud offerings promising both reliability and high availability. The tight coupling between the containers and the Docker daemon proves fatal for the containers' uptime during daemon's unavailability due to either failure or upgrade. We present the design and implementation of HYDRA, which fundamentally isolates the containers from the running daemon. Our evaluation shows that HYDRA imposes only moderate overheads even under load, while achieving much higher container availability.