Research On Container Migration Mechanisms For User Level Fault Tolerance

With the rising of cloud computing,how to migrate services on the faulty nodes to available nodes quickly,flexibly and efficiently becomes an important problem in cloud computing cluster fault-tolerance.Since more and more services are deployed to cloud computing clusters,fault-tolerant mechanisms based on process migration and virtual machine migration hardly meet the requirements of fault-tolerance in cloud computing cluster.Container migration consumes less system overhead than virtual machine migration,and is more flexible and efficient than process migration.Accordingly,a container-based resource allocation modeld and a migration mechanism for user level fault-tolerance are proposed.First of all,fault tolerance of virtualization cloud computing cluster is analyzed.Based on the container virtualization,the physical machines for fault-tolerant are virtualized into a container fault-tolerant pool.And a kind of container fault-tolerant pool resource allocation process under task migration request is given.Complex fault-tolerant resource allocation process in the cloud computing cluster is decomposed into three sub-models: fault-tolerant resource assigning sub-module,fault-tolerant pool management sub-module and container provisioning sub-module.Described by the Markov processes,the three sub-models are integrated into a container fault-tolerant resource allocation integration model.Migration rejection rate and average recovery delay of migration request in the corresponding sub-models are analyzed.Secondly,based on container fault-tolerant resource allocation integration model,sub-models and traditional migration mechanism for fault-tolerant,this paper proposes a container migration mechanism,which contains a container fault-tolerant pool,a container checkpoint-restart method and a container checkpoint file transmission method.Container fault-tolerant pool completes the migrations of the containers for user level,increasing utilization ratio of fault-tolerant resources and cluster fault-tolerance flexibility.Container checkpoint-restart method preserves a full container-process hierarchical relationships and reduces the migration rejection rate.By using the remote memory access technology to achieve checkpoint files transfer,checkpoint file transmission delay is reduced under high I/O overload.Finally,the availability and effectiveness of the proposed model and method are verified by experiments in laboratory environment.