A Method of Self-adaptive Pre-copy Container Checkpoint

Container checkpoint is a kind of backward recovery fault tolerance technology, through which the high availability of container can be achieved. Checkpoint downtime is the key performance indicator of container checkpoint system. A long checkpoint downtime will cause users' perceptual interruption in the deployment of the service of the guest operating system in container, which is difficult to accept for the cloud system offering key services. In order to reduce the downtime of container checkpoint, a method of self-adaptive pre-copy container checkpoint is proposed in this paper. Through several rounds of pre-copy, memory pages of container which will not be frequently modified are copied in advance. Only dirty pages which generate in the previous round of pre-copy are saved in every round of pre-copy by freezing container, which reduces the checkpoint downtime. The number of rounds of pre-copy is adaptively determined by the workload of the guest operating system in container. Prototype of Self-Adaptive Pre-copy Diskless Linux Container Checkpoint (SAPCDLCKPT) is implemented based on Linux container (LXC). The experimental result shows that compared with the existing methods, with the constantly increase of container's memory configuration and different kinds of workloads, SAPCDLCKPT achieves lower checkpoint downtime. The highest decrease ratio of checkpoint downtime reaches to 78.24%.