Live migration of real-time containers

Radhakrishnan, Devikrishna

Permalink

https://hdl.handle.net/2142/120454 Copy

Description

Title

Live migration of real-time containers

Author(s)

Radhakrishnan, Devikrishna

Issue Date

2023-05-03

Director of Research (if dissertation) or Advisor (if thesis)

• Mohan, Sibin
• Bates, Adam

Department of Study

Computer Science

Discipline

Computer Science

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• Iot
• Real-time
• Edge Server
• Container
• Migration
• Criu
• Podman

Language

eng

Abstract

IoT and cyber-physical applications, such as collaborative robots, autonomous transport systems, etc., often have high variability in user location, server loads and network conditions. Such systems can use edge computing for faster and more reliable computational services. With the additional constraints of timing guarantee and predictability brought in by the real-time computational needs of these applications, resource assignment and load balancing must be carried out in real-time. Real-time computations can be run inside containers on the edge servers to facilitate this and to avail the benefits of easy portability. However, it is often difficult to achieve reliable performance with transient load spikes at edge servers as well as the mobile nature of the application itself (e.g., autonomous vehicles offloading compute onto edge servers while on the move). Therefore, to maintain QoS levels, it becomes necessary to migrate the computation to a different, more suitable edge server several times throughout the application runtime. Hence, reducing the live migration time of a container running real-time computations becomes critical. The state-of-the-art container migration approach using CRIU (Checkpoint Restore in Userspace) incurs high migration costs leading to significant application downtimes, which makes it unsuitable for real-time applications. This thesis investigates the container checkpoint and restore pipeline (handled by CRIU) and improves upon it by decoupling fundamentally independent operations enabling their parallel execution, resulting in a real-time ready live migration framework. In addition, the framework achieves fault tolerance against network and operational failures (e.g., in the checkpoint/restore phase) by replicating migration to multiple potential destination edge servers. Evaluation of the framework using two real-time benchmark applications: 360°video and Openface, showcases a reduction in live migration time by ∼[8-65]% across scenarios.

Graduation Semester

2023-05

Type of Resource

Text

Handle URL

https://hdl.handle.net/2142/120454

Copyright and License Information

Copyright 2023 Devikrishna Radhakrishnan

Owning Collections