AngleCut: A Ring-Based Hashing Scheme for Distributed Metadata Management

Jiaxi Liu, Renxuan Wang, Xiaofeng Gao, Xiaochun Yang, Guihai Chen

Abstract: Today’s file systems are required to store PB-scale or even EB-scale data across thousands of servers. Under this scenario, distributed metadata management schemes, which store metadata on a group of metadata servers (MDS’s), are used to alleviate the workload of a single server. However, they present a significant challenge as the group of MDS’s should maintain a high level of metadata locality and load balancing, which are practically contradictory to each other. In this paper we propose a novel and specially designed hashing scheme called AngleCut to partition metadata namespace tree and serve large-scale distributed storage systems. AngleCut first uses a locality preserving hashing (LPH) function to project the namespace tree into linear keyspace, i.e., multiple Chord-like rings. Then we design a history-based allocation strategy to adjust the workload of MDS’s dynamically. The metadata cache mechanism is also adopted in AngleCut to improve the query efficiency. In general, our scheme preserves the metadata locality essentially as well as maintaining high load balancing between MDS’s. The theoretical proof and extensive experiments on trace data exhibit the superiority of AngleCut over the previous literature.