Research On Fast Fault Recovery Strategy Based On Log

With the rise of technologies such as big data and cloud computing,distributed storage systems represented by data centers have become the mainstream.Failure in such system happens a lot,so how to ensure system reliability and optimize the recovery process have extremely important practical significance.Fault recovery technology under copying scenario copies a lot of useless data due to lack of recovery accuracy,which causes bandwidth competition between recovery and I/O service.Resources on faulty node are insufficient to satisfy both,which results in problems such as long recovery time and drastic drop in client service quality.Meantime client services may be wrongly blocked by recovery.Log-based fast failure recovery scheme includes two optimization strategies.Aiming at the problems of large recovery granularity and long recovery time in existing fault recovery schemes,a fine-grained recovery strategy was proposed.By designing and implementing a fine-grained log structure,the client's access history during faults was recorded in detail,and the original logs were analyzed and consolidated.The data interval,in combination with logs,completely eliminates the useless copy of data,effectively alleviates the competition between the recovery process and normal services on the bandwidth resources,and effectively shortens the total time required for recovery.For the client service being blocked by the recovery process and the quality of service drooping,an accurate I/O blocking policy is proposed in conjunction with the log to ensure that the blocking and triggering recovery process occurs only when the client I/O hits the data interval in which the failure period is modified..This strategy can effectively eliminate the problem that the client does not have to be blocked during the previous large-granularity recovery process,and further improve the quality of the client service in the recovery period.Finally,combined with the distributed storage system Ceph,the scheme is implemented,and the performance test is combined with different loads.The experimental results show that the high-precision fault recovery strategy based on the log can effectively eliminate the copying of unnecessary data,and greatly relieve the bandwidth pressure and system resource competition.For different block size random write and sequential write loads,The total recovery time is reduced by an average of 80% and 20%.The log-based accurately client I/O blocking optimization strategy effectively eliminates useless client I/O blocking during recovery.For different block size random write and sequential write type workloads,the client bandwidth in the recovery period averages up to 5 times and 11%.