Abstract: Hadoop adopts replicas as the default data fault tolerance,while this fault tolerance mechanism occupies much storage space and has a low storage efficiency.To solve this problem,this paper proposed an adaptive replacement cache mechanism for fault tolerance in cloud storage,based on analyzing the cache replacement algorithm ARC.When users access the file system,ARCMFF statistics files’ access frequency by maintaining a LRU queue and a LFU queue,and then adds those files highly accessed to the cache for better system performance.In ARCMFF,the majority of files are stored by erasure code,while only few files in the cache are stored by replica.The erasure code is highly encoded and has a higher storage efficiency,resulting that the distributed file system can save amount of storage space.According to series of experiments,we approved that the distributed file systems with ARCMFF can save file storage space greatly,improve storage efficiency and achieve higher reading and writing performance.

Key words: Cloud storage,Data fault tolerance,Hadoop,Replica,Erasure code,ARC