Efficient algorithms for improving the performance of read operations in distributed file system
Talluri Lakshmi Siva Rama Krishna, Thirumalaisamy Ragunathan, Sudheer Kumar Battula · IEEE International Conference on High Performance Computing, Data, and Analytics · 2015
Distributed file systems (DFSs) are used in the modern cloud-based systems to store and process a large amount of data. File sharing semantics is used by the DFS for sharing the data in a consistent manner among the authorized users of the system. The limitation of popular session semantics is that read client processes cannot read the modifications done by a concurrent write client process on the same shared file in the same session. Linearizability semantics followed in the BlobSeer DFS permits the read client processes to read the previous version of a binary large object (blob) while write operation is carried out on that blob concurrently. In this paper, we have proposed a new type of semantics and metric namely Semantics and respectively. Speculative semantics permits the read client processes to read the modifications done by the concurrent write client process in the same session. Currency is the metric used for measuring the performance of the read algorithms. In this paper, we have proposed two new read algorithms based on speculative semantics. We have conducted experiments on Blobseer DFS to measure the performance of these read algorithms and the results obtained indicate that the proposed algorithms perform better than existing read algorithm of the BlobSeer DFS.