| In order to increase the reliability of distributed storage system,it is necessary to recover the damaged data in time.The dominant factor of data recovery is I/O or reading the data from disk.With the increase of storage capacities and CPU performance,I/O becomes the valuable resource and the primary bottleneck in storage systems.The other critical operation in storage system is degraded reads.When a read request arrives,the systematic node storing the requested data may be busy or unavailable.The request is served by calling upon a reconstruction operation to recover the desired data from the remaining nodes.I/O also determines the performance of degraded reads.This paper is working on optimizing the I/O performance in data recovery and degraded reads.Researches are focusing on recovering data from node failures.However,the much more common failure type is sector failures.Sector failures and node failures are both considered in data recovery.Two kinds of sector repair methods are provided based on the structure of RBT codes.For RBT helpers of failed nodes,they can be exactly recovered with node repair.For Non-RBT helpers,I/O can be optimized by calculating the largest common recovery set.Based on the fact that contiguous data are stored on different nodes to maximize the advantage of parallel I/O,the sector repair method provided in this paper also applies for the degraded reads in storage system.Finally,the effectiveness of the provided strategy has been proved through simulation test. |
