Fault Recovery For The On-Board Pipelined Task

As On-Board applications becoming more and more complex, the data dealing by On-Board application also show exponential growth. The multiprocessor platform provides the necessary hardware platform When the single processing unit of computing platform can’t meet the demand of the application for large-scale space borne calculation. Meanwhile, It has to adapt serial program to parallel program for effectively using the parallel platform. The pipelining parallel as one of parallel methods has been widely used in many fields, such as multimedia processing and wireless communications fields. Therefore,Practices have proved that pipelining processing effectively improves the computing performance of space borne applications.At the same time, the cosmic ray with a large number of charged particles such as electrons, protons, and heavy ion particles makes the outer space environment to be a complex environment. And Space borne programs running in such an environment is easy to suffer from a transient fault. So it is necessary to ensure the reliability of the space borne program while we improve computing performance. Fault-tolerant technology is a main and effective means to enhance the reliability of space borne application. So Fault recovery technology as part of the software fault-tolerant technology is very important to guarantee the reliability of the space borne program. Currently, the traditional fault recovery strategy for pipelining parallel program is simply to restart the whole task or system, this coarse-grained recovery strategy to ensure reliability while offering high performance loss.In this paper, we propose a grained recovery policy for pipelining parallel task model separately on the shared memory parallel platform and distributed memory parallel platform, it would ensure the reliability of computing tasks on the basis of reducing the overhead of system for recovering. Then we realize this grained fault recovery mechanism in VxWorks platform and experiment with pipeline JPEG image compression application to prove this fault mechanism works and is efficient.