Cross-Partition Protocols in a Distributed File Service

Zhang, Zheng ; Karamanolis, Christos ; Mahalingam, Mallik ; Muntz, Daniel

HP Development Company

关键词: distributed file system;    distributed namespace;    fault tolerance;    Storage Area Network (SAN);   

RP-ID  :  HPL-2001-129

学科分类：计算机科学（综合）

美国|英语

来源: HP Labs

【 摘 要 】

A number of ongoing research projects follow a partition-based approach in order to achieve high scalability for access to the distributed storage service. These systems maintain a namespace that references objects distributed across multiple locations in the system. Typically, atomic commitment protocols (e.g., 2-phase commit) are used for updating the namespace, in order to guarantee its consistency even in the presence of failures. Atomic commitment protocols are known to impose a high overhead to failure-free execution. In addition, they use conservative recovery processes and may considerably restrict the concurrency of overlapping operations in the system. This report proposes a set of new protocols for the implementation of the fundamental operations in a distributed namespace. The protocols use intention logs to impose a minimal overhead to failure-free execution. They are robust against both communication and host failures, and they use aggressive recovery procedures to re-execute incomplete operations. The proposed protocols are compared with their 2-phase commit counterparts and are shown to be strictly better in all critical performance factors: communication round-trips, synchronous I/O, operation concurrency. 21 Pages