real time kernel: benefits

Posted Nov 18, 2024 6:10 UTC (Mon) by rurban (guest, #96594)
In reply to: real time kernel: benefits by atnot
Parent article: The realtime preemption pull request

I would not call the NASA Space Shuttle program a small player. The shuttle was controlled by 3 dSpace machines, ordinary AMD processors with RtLinux. RtLinux had just more and better drivers than the old simple real-time systems. Eg Gigabit Ethernet with DMA support.

Finally the seperate RtLinux is not needed anymore, it's now maintained upstream.

real time kernel: benefits

Posted Nov 18, 2024 11:06 UTC (Mon) by malmedal (subscriber, #56172) [Link] (1 responses)

> The shuttle was controlled by 3 dSpace machines, ordinary AMD processors with RtLinux.

Uh? The shuttle predated Linux. I believe the modernised cockpit was going to use VxWorks, but the effort was canceled.

real time kernel: benefits

Posted Nov 18, 2024 12:38 UTC (Mon) by excors (subscriber, #95769) [Link]

The Shuttle also predates the company dSPACE, and I can't find any sources indicating any part of the Shuttle program ever used dSPACE products, nor any sources confirming any dSPACE products used RTLinux (though some of their modern products do use a "Linux-based real-time operating system", which I guess might mean PREEMPT_RT). So I think this is a [citation needed].

Apparently the Shuttle's on-board computers were IBM System/4 Pi (based on System/360; the name is a joke about steradians), with a purpose-built OS. The OS was written in assembler and the applications in HAL/S (based on PL/I). Each computer had 424KB memory, could do 450K operations per second, and weighed about 55kg. The Shuttle had four computers in a quadruple-redundant set, plus a fifth running backup flight software developed independently by a different company. (Source: https://klabs.org/DEI/Processor/shuttle/shuttle_primary_c...). Mission Control originally used five IBM System/370 computers (one primary, one backup, three spares) (https://waynehale.wordpress.com/2012/05/28/ground-up-rend...), though I guess that may have been updated over time as it was less safety-critical than the flight computers.

> I believe the modernised cockpit was going to use VxWorks, but the effort was canceled.

Yes, that's discussed in https://ntrs.nasa.gov/api/citations/20070018256/downloads... (which also notes, worryingly, "The project expended a fair amount of time tracking down system problems mostly due to the misuse of pointers and memcpy commands"; and I guess they got a bit carried away with C++ because "It was found that software components utilizing all [the] benefits of [Object Oriented] programming had the worst performance. Several components had to be redesigned and rewritten to fix this problem").