What do determinism, predictability and bounded latency mean for an RTOS, and why do they matter more than raw speed?
Bounded latency means a task that needs the processor is guaranteed to get it within a known time; predictability and determinism mean you know how long each task takes, so you can prove every deadline is met.
The seven key goals of an RTOS all serve one purpose, being able to say in advance that the system will meet its deadlines:
| Goal | Meaning |
|---|---|
| Small latency | Latency and processing time are bounded; every task that needs processor time actually gets it |
| Multi-tasking | More than one task or application is in memory and running at the same time |
| Real-time responsiveness | Inputs and events are handled within a defined short period |
| Predictability | The time several tasks take to execute is known |
| Preemption | A task can be suspended temporarily so a higher-priority task can run |
| Determinism | You know how long things take to process, so deadlines are met |
| Scalability & structured software | Components can be added, scaling from a simple application to one with stacks, drivers and file systems |
The reason this beats raw speed is that designers budget for the worst case. If a motor controller must react within 1 ms and the scheduler can guarantee 50 µs worst-case latency, the design can be verified. A faster processor with an occasional unexplained 5 ms pause cannot be verified, however good its average.
Tip: the question an RTOS answers is not "how fast?" but "how late, at worst?"
Go deeper:
Wikipedia — Real-time computing — hard, firm and soft deadlines, and what "missing a deadline" costs in each.