Performance Evaluation of a Prototype Distributed NFS Server

Abstract

International audienceA high-performance file system is normally a key point for large cluster installations, where hundreds or even thousands of nodes frequently need to manage large volumes of data. While most solutions usually make use of dedicated hardware and/or specific distribution and replication protocols , the NFSP (NFS Parallel) project aims at improving performance within a standard NFS client/server system. In this paper we investigate the possibilities of a replica-tion model for the NFS server which is based on Lasy Release Consistency (LRC). A prototype has been built upon the user-level NFSv2 server and a performance evaluation is carried out

Evaluating the performance of the dNFSP file system

“…This machine is composed of 20 nodes interconnected by Fast Ethernet, where each node features two Pentium III processors at 1 GHz, 1 In the PVFS and dNFSP experiments, a set of 4 nodes has been dedicated to the file server, each one holding one iod. The PVFS manager runs together with the first iod.…”

“…This design allows for increased performance in the case of read operations, because most of the data involved in the complete operation are sent directly from the iods to the clients; however, in the case of writes, the data must be sent from the clients to the metaserver, and thus the original bottleneck remains the same. In order to alleviate this problem, dNFSP [1] has been proposed as a variation of NFSP in which the meta-server is replicated onto several compute nodes. Figure 2 illustrates the new design.…”

“…We rely on the fact that accesses to really inexistent files should not occur (at least not often) in a parallel application. More details on the model are presented on another paper [1].…”

“…Following this approach, in this paper we present a performance evaluation of dNFSP [1], an extension of the standard NFS file server intended for an improved level of performance and scalability on clusters while still maintaining compatibility with the standard NFS clients available on every Unix system. Our main goal is to evaluate the feasability of dNFSP as an alternative for cluster file systems.…”

Impact of I/O Coordination on a NFS-Based Parallel File System with Dynamic Reconfiguration

“…dNFSp -distributed NFS parallel -is an extension of the traditional NFS implementation that distributes the functionality of the server over several machines on the cluster [2]. The idea behind dNFSp is to improve the performance and scalability without sacrificing compatibility with standard NFS clients -thus simplifying setup and maintenance.…”