PDSI SciDAC
The Petascale Data Storage Institute is part of the DOE SciDAC effort. The Sandia lead is Lee Ward with collaborators at Carnegie Mellon University, the University of Michigan, the University of California at Santa Cruz, and DOE labs at NERSC, PNNL, ORNL, LANL, and SNL. The objective is to channel research efforts towards development of solutions to the challenges of data movement at petascale. All results and developed solutions will be made publicly available.
Some research focus areas are:
-
characterization of application performance at scale
-
exploration of protocol and API extensions needed to achieve required performance
-
exploration of novel mechanisms for emerging petascale science requirements
-
exploration of automation for petascale storage system management
The project also provides for a strong education/outreach component, including public lectures and workshops aimed at increasing knowledge and awareness of the impending issues in data storage for petascale computing.
The Sandia Scalable I/O team has developed code to incorporate a light weight tracing capability into the user-level VFS provided by libsysio. Traces of I/O activity from scientific applications running at large scale have been released and are available to researchers to provide input data for simulation tests of new I/O approaches. A workshop was scheduled at SC07 and a colloquium lecture was delivered at the University of New Mexico in January 2008.
libsysio
The Sandia Scalable I/O team is responsible for the design, development, and maintenance of libsysio, an open source library that provides POSIX-like file I/O and name space support for remote file systems from the application program address space. The library was developed for the light weight kernel Catamount and is currently part of the operating system software on Red Storm and other Cray XT3 platforms.
Scaling I/O for Commodity Clusters
In collaboration with Helen Chen at Sandia CA, this project is an investigation of bottlenecks in the Linux NFS client. The objective is to identify and remedy obstacles in the Linux kernel and NFS client code which have a negative impact on streaming write throughput for NFS. It is expected that these changes will allow the RDMA functionality present in pNFS to make better use of the available network bandwidth. A positive result here will open up a freely available, easy to administer, scalable I/O solution in the commodity cluster arena.
Lightweight File System
The Lightweight File System (LWFS) project is a collaboration between
Sandia National Laboratories and the University of New Mexico investigating
the applicability of "lightweight" approaches for I/O on MPP systems.
Top of page
|
