OrangeFS @ CCU Overview
This is an overview of the OrangeFS-related projects currently underway at Coastal Carolina University.
What is OrangeFS?
Orange File System is a branch of the Parallel Virtual File System (PVFS) currently under development at Clemson University. Like PVFS, Orange is a parallel file system designed for use on high end computing (HEC) systems that provides very high performance access to disk storage for parallel applications. OrangeFS is different from PVFS in that features have been developed for OrangeFS that are not presently available in the PVFS main distribution. While PVFS development tends to focus on specific very large systems, Orange considers a number of areas that have not been well supported by PVFS in the past. OrangeFS development tends to be driven by input from users with specific needs.
We have teamed up with two groups at Clemson University (PARL and Omnibond) in order to help them provide some additional features in OrangeFS. We were awarded a 15-month subcontract to work on the requested features. The following describes the nature of this work.
Statement of Work
We propose to work with the Clemson Orange FS team to develop a fully redundant version of the code. This will include not only redundant data, but also redundant metadata, attributes, and directory structure. The proposed modification will allow attributes on files to control how many redundant copies of data will be stored and when those copies are updated (never, on immutable, on close, on sync, continuously). Files opened in direct mode will have operations that trigger replication block until the replication is complete. Otherwise replication can happen asynchronously provided semantics are maintained. Attributes on directories serve as defaults for files and directories created in them. Directory and metadata replication is controlled by a separate set of attributes. File system defaults are specified in the file system configuration database. Tools will be provided to see the state of these various attributes, and the current status of replication in the file. Counters will be provided to monitor various actions through the existing management interface.
Dr. Jones and students (Brian Atkinson) working with him at CCU will work with the Clemson OrangeFS team to divide the project into manageable tasks, and select those tasks appropriate for CCU. Dr. Jones would then work on those tasks as agreed upon for the duration of the funding. It may be prudent to set intermediate goals that represent only part of the work described above, and the scope of work assigned to CCU may need to be adjusted as time and funding permits. Dr. Jones will work with the Clemson team and his students to report on the status of the project, will prepare papers and presentations about the project for relevant conferences including but no limited to SC11 and SC12. As a final report, documentation on all completed work will be provided to the Clemson team. All code and documentation will become part of the PVFS distribution, and open source code licensed under LGPL. Dr. Jones and his students will all be listed as contributors to the project.