A high-performance adaptive strategy of container checkpoint based on pre-replication

Abstract

During the implementation of the container checkpoint strategy, checkpoint downtime is a pivotal performance indicator. Shorter downtime is especially important for systems that provide critical services. To reduce the checkpoint downtime, an adaptive pre-replication checkpoint strategy named APR-CKPOT is proposed in this paper. Through several rounds of pre-replication, the infrequently modified container memory pages are preferentially copied. The dirty pages generated in the previous round of Pre-Replication are saved in each round of pre-replication. The number of pre-replication checkpoints is adaptively determined by the workload of the user’s operating system in the container. The coordination between fault-tolerance service capabilities and performance of the container can be achieved, and the downtime of the checkpoint can be reduced, which is verified by the given experimental results based on Docker container system.