distributed system, Hadoop, Blobseer, concurrency, speculation
File-sharing semantics is used by the file systems for sharing data among concurrent client processes in a consistent manner. Session semantics is a widely used file-sharing semantics in Distributed File Systems (DFSs). The main disadvantage of session semantics is that writes to an open file are visible to the concurrent client processes only during their next session. Recently, “linearizability semantics” was introduced in BlobSeer DFS, in which a Read Client Process (RCP) can read only a previous version of a binary large object (blob), while update operations are carried out on that blob in a concurrent manner. In this paper, we propose a new type of file-sharing semantics, namely “speculative semantics”, which permits writes to an open file to be visible to other concurrent processes provided that data consistency is not affected. In addition, we propose a new read algorithm for DFSs based on speculative semantics and a new performance measurement metric called Currency. The experimental results obtained using BlobSeer DFS indicate that the proposed read algorithm performs better than the existing read algorithm of BlobSeer DFS.
Tsinghua University Press
Talluri Lakshmi Siva Rama Krishna, Thirumalaisamy Ragunathan, Sudheer Kumar Battula. Improving Performance of a Distributed File System Using a Speculative Semantics-Based Algorithm. Tsinghua Science and Technology 2015, 20(6): 583-593.