CDMCR: multi-level fault-tolerant system for distributed applications in cloud

摘要

Cloud provides users with a new model of utilizing the computing infrastructure with the ability to perform parallel and distributed computations using elastic virtual cluster. However, the multi-level and complex features make cloud computing system more prone to failure. In this paper, we present a multi-level fault-tolerant system for distributed applications in cloud named Distributed-application oriented Multi-level Checkpoint/Restart for Cloud (CDMCR). The CDMCR system backups the complete state of applications periodically with a snapshot-based distributed checkpointing protocol, including file system state. Thus, we cannot only recover processes but also rollback data. A multi-level recovery strategy is proposed, which includes process-level recovery, virtual machine recreation, and host rescheduling, enabling comprehensive and efficient fault tolerance for different components in cloud. We deploy CDMCR as PaaS, so that users can be liberated from node management and system configuration and get access to fault-tolerant service conveniently. We have implemented this system based on the Xen virtualization platform and the OpenNebula cloud platform. Experiments on the prototype demonstrate the correctness of the system. Analysis shows that CDMCR does not cause message loss or data loss, and the backup time remains nearly constant as the number of nodes increases on virtual cluster. Copyright (c) 2015 John Wiley & Sons, Ltd.