Container migration for mobile services

Nowadays, the development of edge computing technologies enables the deployment of time-sensitive and large-data-volume services on infrastructure, which significantly promotes the development of the applications like connected vehicles and the intelligent industry. Together with the development of mobile communication technologies(e.g., 5G), edge computing architecture is gaining more and more traction, leading people to imagine the possibility of deploying time-critical mobile services on edge to reduce computational power requirements on mobile devices, especially in application scenarios like autonomous driving or unmanned aerial vehicles (UAVs). However, in actual practice, the deployment of time-critical mobile services on the edge server still has many challenges, especially the communication latency. When a mobile client moves, the communication latency between service and client will increase as the communication distance increases, which is mainly related to the unpredictable network congestion. In this case, the communication latency will no longer be guaranteed when the mobile client reaches a specific distance with respect to the fixed edge service. The most feasible solution to this problem is to enable the mobility of mobile services among the edge network, making the mobile services can always be reachable in the edge server close to the mobile client. Since edge services are normally containerized, service migration in the edge network could be generalized into container migration. Even though compared to the VM migration, container migration is very lightweight and flexible, the migration costs and benefits still need careful balancing in time-critical services migration. In the work of this thesis, the study on container migration technology will focus on the following two aspects: container migration realization and migration cost modeling. In the part of container migration realization, general requirements and constraints for container migration will be discussed. And in practice, Podman container will be used to containerize services since it has better integration with CRIU, a key tool to realize container migration on Linux machines. In the part of migration cost modeling, the container migration cost will be studied from the aspect of service downtime and migration resource usage (including CPU, memory, storage, and network usage). An overall cost model will be proposed in order to define different migration policies for different kinds of edge applications. Since container migration is a new study topic in the edge computing area, the study in this thesis will focus more on the low-layer migration implementation and performance. Future work could concentrate on the study of high-layer orchestration or the discussion of the extreme conditions on container migration, completing the technical stack and pushing the adoption of this technique in real applications.