Research On Checkpointing And Rollback Recovery Fault-tolerant Techniques For Mobile Computing Environment

With the rapid development of the high performance computing, internet, wirelesscommunication, distributed computing, pervasive computing and cloud computing fields, thedevelopment of the mobile computing technology has been greatly promoted. Compared tothe traditional wired distributed computing environment, mobile computing system has broadapplication foreground due to its various features, such as celerity of setting, autonomy,dynamic mobility of the node, flexibility of topological structure and equivalence. However,the process failure event probability of mobile computing environment is greater than that ofthe traditional wired distributed computing system. The checkpoint and rollback recoveryfault tolerant mechanism for the wired distributed computing system is inappropriate for themobile computing environment. Therefore, it is meaningful to design an appropriate efficientcheckpoint and rollback recovery fault tolerant mechanism for mobile computing system.According to the current researches and existent problems of the fault tolerant schemes, thisdissertation is mainly focus on the checkpoint and rollback recovery fault tolerant techniquesfor mobile computing environment. The contents of the dissertation are:(1) The storage and maintenance of the process log for mobile computing is studied. Alow overhead process event logging mechanism has been proposed based on m-MSS-mmodel and the piece-wise deterministic execution model assumptions. In the proposal, thecheckpoint, the event log and the happened-before relation of the mobile host process in thelocal cell are stored and managed by the mobile support station uniformly. The checkpointand the event log in the form of the determinant are recorded in the single-dimensional array.Specifically, the happened-before relation among the event experienced by the process isindicated by the sequence of the array. The process log is recorded in the high-speed memoryof the mobile support station synchronously, and the log is flushed into the reliable storageupon the special event asynchronously.(2) The rollback recovery of the failure process for mobile computing is studied. Arollback recovery mechanism of the failure process has been proposed according to theproposed logging mechanism. The recovery and logging mechanism constitute the checkpointand rollback recovery fault tolerant mechanism based on the event logging. In the case of the complete fault tolerant log, the failure process can implement independent consistent rollbackrecovery. In the case of the incomplete fault tolerant log, the processes in the local cell requireimplementing a coordinate consistent rollback recovery.(3) The handoff maintenance of the process recovery information for mobile computingis studied. To balance the system performance during the failure-free execution phase and therollback recovery phase after the failure event, a weak handoff management mechanism basedon the redundant information partition is proposed. Logically, the fault tolerant information ofa mobile host is partitioned into two parts, including the kernel and un-kernel parts. When amobile host incurs a handoff, the different fault tolerant logs are managed in times anddifferent styles. Finally, the constraint for determining the kernel and un-kernel part amountsof the process is derived in this dissertation.(4) The determination of the process checkpoint interval in the fault tolerant mechanismis studied. For Passion failure distribution, an equidistant checkpoint interval based onLaplace Transformation is proposed to ensure the entire performance of the fault tolerantmechanism for mobile computing. For the other arbitrary failure distribution, the averageprocess computation effective rate of the fault tolerant mechanism is derived based on asimple checkpointing timing method. The general constraint of the optimality is presentedaccording to the average process computation effective rate, and a general checkpointscheduling algorithm is developed to perform a qausi-optimal process checkpoint sequence.The performance analysis shows that the proposed fault tolerant mechanism for mobilecomputing is considerable in various aspects, such as the recording and maintenance of theprocess checkpoints and logs, the recording and maintenance of happened-before relationamong the processes, the determination of optimal checkpoint interval, the handoffmaintenance of the fault tolerant information and the independence of the rollback recoveryprocess. The research result of this dissertation is an efficient fault tolerant mechanism for thereliability of mobile computing.