Safety-critical systems [LWN.net]

Safety-critical systems

Posted Nov 21, 2008 9:05 UTC (Fri) by epa (subscriber, #39769)
In reply to: Linux Guru Reiser Seeks New Murder Trial (Wired) by jd
Parent article: Linux Guru Reiser Seeks New Murder Trial (Wired)

There is no way Linux is reliable enough for safety-critical systems like surgical lasers or flight control systems _anyway_. Hans Reiser's case doesn't change that.

(Log in to post comments)

Safety-critical systems

Posted Nov 21, 2008 10:30 UTC (Fri) by tialaramex (subscriber, #21167) [Link]

Of course the same type of media "sources" that would introduce confusion about whether somehow Linux is "the work of a murderer" because one of many thousands of contributors killed someone wouldn't consider it important to distinguish between...

- The aeroplane runs Linux on an embedded PC which provides 64 channels of streaming music to passengers. If it fails passengers will watch a movie.

- The aeroplane runs Linux on a flight-related but non-essential system. If it fails the pilot will curse it and log it as a fault, but there's a million-to-one chance of this distraction contributing to an accident.

- The plane's engines run on a mixture of fairy dust and Puppy Linux. A bug in the IRC client will cause the wings to fall off and send your plane nose-diving into a mountain.

And it's not as though they're picking on us. When car manufacturers look at using Windows embedded, the same sources will happily assume that somehow Windows will control the engine and steering rather than, as is closer to reality, a DVD player and the stereo.

Safety-critical systems

Posted Nov 24, 2008 16:31 UTC (Mon) by AJWM (guest, #15888) [Link]

It's not a matter of reliability, as such. For some such applications and configurations, it might well be.

It's a matter of certifiable reliability. I'm not as up on the FDA requirements, but certainly FAA has some pretty strict requirements for documenting the process by which flight control software was developed, including documenting the specifications and so on, steps that harken back to the old waterfall method, quite different from typical FLOSS methodology. (It's slow as hell, but it does - when followed properly - produce reliable software (which then runs on hardware two generations old)).

That doesn't apply to experimental aircraft, of course -- wouldn't surprise me if there are some homebuilts using Linux for flight control software, so long as there's a manual override.