Hacker News new | past | comments | ask | show | jobs | submit login

It would be fun to try to design a system that can work around all kinds of failures, such as being able re-purpose one subsystem to take over from another one.

For example, have all the digital electronics be reprogrammable gate arrays.




Relative simplicity is also a key ingredient to the success of these probes. The computers onboard are completely understandable by a single person. The software stack doesn’t have so many layers of abstraction that you are disconnected from what the hardware is actually doing… in fact the opposite is true.


Redundancy is probably the best option. That’s how the Voyager is still working after one of the computers failed.


Redundancy is a very good option, but the "right" level of modularization helps too. You probably don't want to make everything out of FPGAs, but if your individual modules can be flexibly rewired on the fly and they are granular enough in their functionality to build new things, that adds yet another level of redundancy, as you might be able to replace a broken thing with the right combination of working modules.


Redundancy is the straightforward way, two of everything. That still means two failures can cripple the machine. I'm talking about going beyond that. I.e if you need A and B and have modules X, Y, and Z, then any one of XYZ can fail and you still have full functionality. I.e. you can have more redundancy with fewer modules.


Reminds me of something I got really invested in about a decade ago which is Dave Ackley's work on "robust first computing". The example at the end of the sorting algorithm that can reconstitute itself when you blow half of the machine up is still amazing to me.

https://youtu.be/lbgzXndaNKk


Not possible in reality. You'll inevitably have single points of failure in the control system of any spacecraft, even with redundancy, rerouting, or tech that is robust to individual gate failures (erasure coding, holographic storage, neural networks etc).


One massive FPGA? In the future: nanobots that can reassemble to a better design. Maybe you don’t even need to complete the design, just ship the nanobots and do a 0-day patch when they reach their target.


Or they will deconstruct the shop in flight. New version of rapid unscheduled disassembly or RUD, normally reserved for crashes and flight failures.


They should make a movie.


I wonder how FPGAs handle cosmic radiation.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: