Low-Overhead Protocols for Fault-Tolerant File Sharing

In this paper, we quantify the adverse effect of file sharing on the performance of reliable distributed applications. We demonstrate that file sharing incurs significant overhead, which is likely to triple over the next five years. We present a novel approach that eliminates this overhead. Our approach (1) tracks causal dependencies resulting from file sharing using determinants, (2) efficiently replicates the determinants in the volatile memory of agents to ensure their availability during recovery, and (3) reproduces during recovery the interactions with the file server as well as the file data lost in a failure. Our approach allows agents to exchange files directly, without first saving the files on disks at the server. As a consequence, the cost of supporting file sharing and message passing in reliable distributed applications become virtually identical. The result is a simple, uniform approach, which can provide low-overhead fault-tolerance to applications in which communicatio...

• Type: Conference
• Year: 1998
• Where: ICDCS
• Authors: Lorenzo Alvisi, Sriram Rao, Harrick M. Vin
• Keywords: Distributed And Parallel Computing, File Sharing, File Sharing Incurs, ICDCS 1998, Reliable Distributed Applications