Self-Healing Computers For NASA Spacecraft 70
Roland Piquepaille writes "As you can guess, hardwired computer systems are much faster than general-purpose ones because they are designed to do a single task. But when they fail, they need to be totally reconfigured. This can be just a costly problem in a lab on Earth, but it can be vital in space. This is why a University of Arizona (UA) team is working with NASA to design self-healing computer systems for spacecraft. The UA engineers are working on hybrid hardware/software systems using Field Programmable Gate Arrays (FPGAs) to develop these reconfigurable processing systems. As the lead researcher said, 'Our objective is to go beyond predicting a fault to using a self-healing system to fix the predicted fault before it occurs.'"
Not new (Score:5, Informative)
Re:Beauty in Simplicity (Score:5, Informative)
Reconfigurable Computing, Fault Tolerance (Score:2, Informative)
Re:The future of pr0n! (Score:1, Informative)
Roland the Plogger again (Score:4, Informative)
It's Roland the Plogger again, pushing his ad-laden blog. The actual research summary is here [arizona.edu]. The real paper won't be out until July.
This isn't new. JPL has been trying various levels of self-healing for years.
The original article describes a cluster of five machines, set up so that if one fails, others take over tasks running on the failed machine. That's what the better server management systems do. I went to a talk last week by Amazon's CTO, and he described how their platform does that.
The project web site makes things clearer. There are two levels of recovery. The upper level works like cluster fallover. The lower level tries to reconfigure the FPGAs to use different cells in the FPGA to work around faults. That's likely to be a delicate process; you'd need substantial on-chip test resources to reliably do gate-level fault isolation on an FPGA that's been hit hard by a cosmic ray. It's not clear how fine-grained this is; this may be more like having multiple units like GPU shaders replicated in an FPGA, with the ability to turn off the failed ones. Sort of like the way Sony ships PS3 machines with eight Cell processors, at least seven of which work.
The available info isn't enough to tell whether this is a good idea or not. About typical for Roland the Plogger.