Parallel wide-area file system testing
This image shows the locally provided
Lustre-WAN storage pool and the connections between it and the remote
storage pools provided by the other organizations participating in the
Lustre-WAN development and evaluation. The image also shows the Data
Capacitor WAN clients. The Lustre-WAN clients, from the same and other
TeraGrid organizations, are not shown.
The collaborative nature of the TeraGrid stresses the need for unified
and global name spaces across participating centers and systems.
Currently this is provided by a WAN file system exported by SDSC using
the multicluster capabilities of GPFS, and NCAR's computational resource
frost provides access to this file system on the frost front-end nodes.
Although this is a production file system that provides a modicum of the
desired features, there are multiple technical and licensing issues with a
GPFS controlled central pool of storage which limits performance and client
support, and thus restricts broad client access across the TeraGrid.
There are currently two TeraGrid-based projects to evaluate the capabilities of the Lustre file system to provide a similar service: the Data Capacitor WAN file system from Indiana Univerisity that is similar in design to GPFS-WAN, and the Lustre-WAN project that aims to provide a kerberos-secured global namespace which aggregates storage pools from across the TeraGrid. Over the past several years, CISL engineers in the ReSET group have deployed and evaluated Lustre to continue to track its development and develop the operational expertise to support production implementations, and this work has allowed us to participate in the ongoing TeraGrid evaluations of Lustre over the WAN.
Over the course of FY2008, ReSET engineers participated in the planning and organization of the Lustre-WAN project, as well as work to evaluate the underlying technologies and prepare local resources for inclusion in the Lustre-WAN file system.
During FY2008, CISL's ReSET engineers:
- Worked with CISL and UCAR security staff to provide a kerberos realm to the TeraGrid systems
- Evaluated Lustre kerberos support
- Configured and allocated local storage resources for integration with the larger Lustre-WAN efforts, which consists of a dual-processor system with an attached 24 TB of storage
In FY2009, CISL will continue to deploy the 24 TB of local storage as part of Lustre-WAN, as well as track and evaluate other Lustre-WAN developments. There are also plans to more fully evaluate the Data Capacitor WAN file system by implementing the required custom patches on dedicated local clients.
This work is supported by NSF Core funding.