Dynamic hashing: Adaptive metadata management for petabyte-scale file systems

Weijia Li,Wei Xue,Jiwu Shu,Weimin Zheng
2006-01-01
Abstract:In a petabyte-scale file system, metadata access performance and scalability will significantly affect the whole system's performance and scalability. We present a new approach called Dynamic Hashing (DH) for metadata management. DH introduces the RELAB (RElative LoAd Balancing) strategy to adjust the metadata distribution when the workload changes dynamically. Elasticity strategy is proposed to support the MDS cluster changes. WLM (Whole Lifecycle Management) strategy is presented to find hot spots in the file system efficiently and reclaim replicas for these hot spots when necessary. DH combines these strategies and Lazy Policies borrowed from the Lazy Hybrid (LH) metadata management together to form an adaptive, high-performance and scalable metadata management technique.
What problem does this paper attempt to address?