The Optimization Of Cross-rack Data Repair Technology For Distributed Storage Systems Based On Ceph

With the generation of massive data,erasure codes become a fault-tolerant solution favored by distributed storage systems.However,erasure codes increase the overhead of disk I/O and network resources when reconstructing lost data.And the repair time is long and the repair efficiency is low.Besides,modern distributed storage systems have complex network structures and usually organize nodes in racks.However,the cross-rack bandwidth is more intense than that within the rack,so the cross-rack bandwidth resources are scarce.Therefore,the focus of this paper is how to recover the unavailable data efficiently and with low overhead when failures occur for the cross-rack network.The paper mainly optimizes the repair process from two aspects:data source selection and data transmission.The main contributions are as follows:1)a data source selection algorithm based on network delay is proposed.The algorithm is used to select the set of data sources when repairing lost data.It select enough nodes with the smallest network delay as data sources to transmit data each time.thereby it can reduce repair time.2)A distributed prototype system based on erasure code is designed.The system implements basic functions such as reading and writing files,heartbeat communication and erasure codes.It verifies the effectiveness of data source selection algorithm based on network delay.The implementation shows that the algorithm can effectively reduce the data recovery time by 36.7%.3)Optimization of data repair process for Ceph system.In Ceph system,the hierarchical decoding algorithm is used to optimize the decoding process of erasure code.The algorithm splits the repair operation into multiple partial decoding operations,and data is transmitted in parallel in different racks,effectively reducing the occupation of the cross-rack network.Experiments show that the hierarchical decoding algorithm can effectively reduce data recovery time by 23.8%.The data source selection algorithrm based on network delay and the hierarchical decoding algorithm are orthogonal.The two algorithms can be combined to recover data,or they can be used separately.