This is the new site of NFSp project. The old NFSp site is no longer maintained.
The new version of NFSp will try to integrate several hot features such as distributed meta-servers, high-level I/O scheduler and fault tolerance. Our main goal is still the same: the client must be able to work, unmodified, that is the server must work as if it were a traditional NFS Server. A production version is currently scheduled for mid-summer 2006.
NFSp stands for Parallel NFS. The main idea of this project is to use several nodes as I/O backend for one server, which is
seen by clients like a regular NFS. This means that NFSp is a plug-in replacement for any currently used server, at the same time,
thanks to the I/O nodes used for actual I/O operations it provides very good performance.
This project was the subject of Pierre Lombard's PhD thesis in 2001. The implementation which accompanied the dissertation supported NFS version 2 and was available as standalone application or a patch for the Linux 2.4 kernel.
The first implementation has brought several concepts to apply to a NFS server such as data striping, I/O servers redundancy, .... . Yet, the main drawback this version had was
there was only one meta-data server: since it acted as a central point through which every request had to go, it was indeed a scalability limitation (especially for write performances).
In 2003, Rafael Avila started his PhD on a new distributed version, called distributed NFSp.
The dNFSp project takes the NFSp concept to the next level. It's main goal is to provide better performance to applications exploring both read and write operations, since the last one is not tackled by the original model. The basis for the proposed system is a distributed metadata management model, which allows for better scalability and reduces the computational cost on the original NFSp metaserver, and also a relaxed mechanism for maintaining metadata coherence based on LRC (Lazy Release Consistency), which enables the distribution of the service without incurring in costly data synchronisation operations.
|10 July 2006||Metaserver Locality and Scalability in a Distributed NFS||VECPAR 2006, Rio de Janeiro, Brazil. submitted version (final version not freely available).|
|9 May 2005||Evaluating the Performance of the dNFSP File System||CCGRID 2005, Cardiff, UK. submitted version (final version not freely available).|
|27 October 2004||Performance Evaluation of a Prototype Distributed NFS Server||SBAC 2004, Foz do Iguaco, Brazil. submitted version (final version not freely available).|
|15 November 2003||Storage and Transfer Solution Between Clusters||"Supercomputing 2003, poster for the demonstrations held at the INRIA booth 3315.|
|10 September 2003||Improving the Performance of a Distributed NFS implementation||PPAM 2003, Poland. submitted papper (final version not freely available).|
|10 September 2003||Distributed File Systems for Clusters and Grids||PPAM 2003, Poland. Special Session "Large Scale Scientific Computations (final version not freely available).|
|27 June 2003||Parallel Data Transfer Solution for Grids||GGF'08, Seattle US (powerpoint presentation).|
|19 April 2002||nfsp: A Distributed NFS server for Clusters of Workstations||IPDPS 2002, submitted version (final version not freely available)
power point presentation available here.
|December 2003||NFSp: Une solution de Stockage Distribue pour Architectures Grande Echelle||Pierre Lombard's PhD thesis.|
|21 June 2003||Systeme de fichiers pour grilles de calcul||Olivier Valentin's Master thesis (French DEA).|
|25 June 2002||Composition de service de donnees et de meta-donnees...||Adrien Lebre's Master thesis (French DEA).|
|12 April 2002||Serveur NFS distribue pour grappe de PCs||RENPAR'14-SYMPA-ASF-8, submitted paper (final version not freely available).|
If you have any questions or comments feel free to e-mail us.
If you think your question is related to a peculiar NFSp component, feel free to contact directly the person in charge:
- Yves Denneulin - leader and official coordinator of data researches led inside the MESCAL team.
- Pierre Lombard - project founder (currently working for ICATIS company but still involved in the project).
- Adrien Lèbre - coordinator of the NFSp activities, schedule strategy designer/developer, project founder of aIOLi.
- Olivier Valentin - developer for the 2.4 kernel implementation (currently involved in the Gedeon project).
- Carlos-Jaime Barrios Hernandez - current gxFer maintainer (working on a PhD about efficient parallel data transfers between clusters).
- Christian Guinet - gxFer co-founder and maintainer till 2003 (currently working for eServGlobal company).
- Rafael Avila - project founder of distributed NFSp
- Everton Herman - dNFSp maintainer (makes a master degree on NFSp fault tolerant mechanisms).
- Danilo Fukuda Conrad, Francieli Zanon Boito, Rodrigo Virote Kassick- dNFSP developers.