Skip to Main Content
Metadata management is critical to distributed file system. Many large-scale storage systems such as HDFS involve the decoupling of metadata management from file data access. In such architecture, a single master server manages all metadata, while a number of data servers store file data. This architecture can't meet the exponentially increased storage demand in cloud computing, as the single master server may become a performance bottleneck. This paper presents a metadata management scheme based on HDFS for cloud computing. Our scheme employs multiple Namenodes, and divides the metadata into “buckets” which can be dynamically migrated among Namenodes according to system workloads. To maintain reliability, metadata is replicated in different Namenodes with log replication technology, and Paxos algorithm is adopted to keep replication consistency. Besides, caching and prefetching technology is integrated to improve system performance.