Research On Key Technology Of Coordinated Rollback-recovery Protocols In Cloud Platform

Cloud computing, which based on the tradition technology and adding newideas, by using the cluster of commercial computer to deal with large amounts ofdata, has become a popular computing model.Howerver, the fault tolerance of cloudcomputing system also becomes a bottleneck increasingly,so there is an urgent needto improve the fault tolerance of the system.The fault-tolerant technology that cloud computing involves is not a newtechnology, include the cooperative checkpoint and message log which all have beenused widely. But facing with cloud computing, these fault-tolerant technology areinsufficient,because they are implemented for virtual machine instances in the mostof cloud platforms.Thence,there exist the need for further improvements to suit thespecial architecture of cloud computing, to realize the fault tolerance of overallsystem in the cloud computing environment.This cooperative rollback recovery techniques of cloud platform which realizedin this paper periodicly checkpoint the state of virtual machines,and synchronize allvirtual machines to avoid orphan message,and through the message drivenmechanism eliminates halfway message,which ensure the checkpoints are globalconsistent checkpoint. When the error occurs, it can quickly detect the error, andperform rollback recovery. Under normal conditions, the virtual machine instancesassigned to different users are independent, so rollback recovery of all virtualmachine instances may lead to a large number of computation loss.In order to reducethe number of virtual machine which rollback recovery, coordinated checkpointalgorithm uses the log.When the error occurs,it only recover the virtual machineswhich depend on the failed virtual machine. Different from the traditionalfault-tolerant technology, the fault-tolerant platform of this paper is transparent forthe specific application and OS, and modules are implemented in a virtual machineprivileged domain in addition to the control module in the cloud platformmanagement server,without need to modifying the application software and OS.This paper studies various types of cloud platforms, and chooses open sourcesoftware CloudStack and XenServer to build IaaS cloud platform,and tests therollback recovery systems that we design and develop. The results of test show thatthe half cooperative rollback recovery protocols provide fault tolerance for cloudplatforms, while reducing waiting time to restore a virtual machine to run, andreducing the number of virtual machines which rollback recovery.