FLARE

Formal Linear Algebra Recovery Environment

John Gunnels (UT-Austin)
Daniel Katz (JPL)
Enrique Quintana-Orti (UJI-Spain)
Robert van de Geijn (UT-Austin)

To be kept informed sign the FLAME guest book



Overview

Within NASA's High Performance Computing and Communications Program, the Remote Exploration and Experimentation (REE) project at the Jet Propulsion Laboratory aims to enable a new type of scientific investigation by taking commercial supercomputing technology into space. Transferring such computational power to space will enable highly-autonomous, flexible missions with substantial on-board analysis capability, mitigating control latency issues due to fundamental light-time delays, as well as inevitable bandwidth limitations in the link between spacecraft and ground stations. To do this, REE does not intend to develop a new computational platform, but rather to define and demonstrate a process for rapidly transferring commercial high-performance computing technology to ultra-low power, fault-tolerant architectures for space. The traditional method for protecting spacecraft components against faults caused by natural galactic cosmic rays and energetic protons has been radiation-hardening. However, radiation-hardening lowers the clock speed and may increase the required power of a component. Even worse, the time needed to design and bring a radiation-hardened component into production guarantees that it will be outdated when it is ready for use in space. Furthermore, it has a high cost which must be spread over a small number of customers. Typically, at any given time, radiation-hardened components have a power:performance ratio that is an order of magnitude lower, and a cost that is several orders of magnitude higher than contemporary commodity off-the-shelf (COTS) components. The REE project is therefore attempting to use COTS components in space and handling, via software, the faults that will occur. As part of the Formal Linear Algebra Recovery Environment (FLARE) project we are pursuing the theory and practice of providing algorithmic fault--tolerance through linear algebra libraries.

Related Publications

For related publications, see the FLAME publication web page.

Related Projects


Back to FLAME page

This web page is maintained by
Robert van de Geijn
flame@cs.utexas.edu
Last Updated: Dec. 19, 2000