PARALLEL DATA LAB 

PDL Abstract

Examining Semantics In Multi-Protocol Network File Systems

Carnegie Mellon University Technical Report CMU-CS-02-103, January, 2002.

Edward P. A. Hogan1, Garth A. Gibson2, Gregory R. Ganger3

1 Panasas Inc.
2 School of Computer Science
3 Department of Electrical and Computer Engineering

Information Networking Institute
Carnegie Mellon University
Pittsburgh, PA 15213

http://www.pdl.cmu.edu/

Network file systems provide a robust tool that can be used by many physically dispersed clients. They provide clients with a means of permanent storage and communication. In order to exploit the resources available on a network file system server, a client must use the protocol of the server's file system. Although the goal of any protocol is to guarantee that the client and server can communicate, the introduction of new protocols divides clients into incompatible sets. Soon clients can no longer cooperate and share because they are using different protocols. In addition, each network file system is constructed with a different set of semantics. The result is that it is increasingly difficult to provide a single storage solution that supports all of these clients. Although difficult, it is extremely desirable to build a multi-protocol network file system, that is, a storage solution that can be used simultaneously by clients of different protocols and semantic sets. A semantic mismatch is a major complexity in building a multi-protocol network file system. These are situations that arise when the normal behavior of a server, expected by a client using a particular semantic set, does not occur because of the effects of a client from a separate semantic set. To achieve the goal of building a multi-protocol file system, the file system semantic sets of the targeted file systems must be carefully examined to determine where semantic mismatches will occur. Next, the possible means of resolving a semantic mismatch can be analyzed for their particular trade-offs. Finally, data from file system traces can be used to determine the frequency of possible semantic mismatches. The data collected from the file system traces, when examined in the context of a cost-benefit analysis, can provide designers of multi-protocol network file systems with important information for examining and resolving semantic differences.

FULL PAPER: pdf / postscript