Safety-critical realtime with Linux

But these discussions about numbers are indeed pointless without looking at the concrete closed loop on a concrete system. I mean, I could tell you that we already measured <1 µs timer interrupt latency in a bare-metal Jailhouse cell besides a loaded Linux on a nice Intel box. But that did not measure any real I/O. Depending on your hardware, the I/O path may be affected by parallel activity of other partitions in the system or hardware-inherent significant jitters or even indeterminism.

So the key takeaways should be:
o Jailhouse allows to remove itself deterministically from the time-critical path - on suitable systems - or at least adds only deterministic delays on the rest
o Hardware plays the major role in what can actually be achieved, deterministically and safely

Specifically the latter point is what makes it challenging right now to certify any software, OSes, hypervisors, Open Source or proprietary on today's complex multicore boxes for safety critical purposes.