NASA's future spaceborne science missions will require supercomputing capabilities for both near earth and deep space exploration. Limited downlink bandwidth and excessive round trip communication delays limit the capabilities and science value of missions which rely on terrestrial supercomputing resources. Projects such as the Gamma ray Large Area Space Telescope (GLAST), the Next Generation Space Telescope (NGST), and autonomous rovers being designed for Mars exploration in the next millenniumwill require onboard supercomputing capabilities to either enable or to greatly enhance their baseline missions. The difficulty encountered by these projects is that radiation-hardened components are both extremely expensive and lag several generationsbehind the commercial state of the art. The goal of the Remote Exploration and Experimentation (REE) project, part of NASA's HPCC program, is to migrate ground-based commercial supercomputing technology into space in a timely and cost-effective manner.Reaching this goal will enable new classes of science missions and make feasible the next major thrust in space exploration. The approach being taken on the REE project is to exploit a comprehensive architecture strategy to enable direct insertion of theprevailing generation of state of the art commercial (hardware/software) components in future space systems. The use of state of the art commercial hardware, coupled with a software-based fault tolerance strategy will allow high throughput computationeven in the presence of relatively high rates of radiation-induced transient upsets as well as in the presence of permanent faults. Utilization of commercial state of the art software components allows the use of standard software development toolsincluding compilers and debuggers and will simplify and speed the development and porting of application codes to the REE computer and their insertion into space-based systems. In this paper, we outline the overall project plan and status, and review thearchitecture of the First Generation Testbed, which is currently being fabricated by a team whose members include Sanders, Caltech JPL, The University of Illinois, and MPI Software Technologies.
展开▼
