An improved method for HDFS replica recovery
Sicheng Liu, Dongxu Duan · 2020
In a cloud storage system, a failure of the data server can lead to increased traffic to other normal servers, increasing the burden on these nodes. There are many defects in the replica recovery method adopted by HDFS, the distributed file system of Hadoop. Such as the concentration of the data transfer, load imbalance. To address this issue, we put forward a replica recovery algorithm based on SVM (RRAS). When DataNodes fail, HDFS choose the source nodes and destination nodes reasonably, and recover the hot data block first. Experimental results show that compared with the default method of HDFS, our method can effectively reduce the time of copy recovery and realize the load balancing of the HDFS cluster.