real-time response accurate to 8 milliseconds range. I am not that familiar with embedded software, but does that value really difficult to achieve?
Bear in mind that real time performance and throughput are perpetually at odds with each other. Long story short, scheduling things such that everything happens in a tight time window means you need to leave gaps in the timeline.
This tradeoff is variable so you can loosen requirements for running something like a rover where there isn't really all that much need for tight realtime control, it's not trying to achieve micron positioning as it drives around. That means you can use a more efficient CPU and lower your idle current which is probably the real constraint here.
8 ms isn't hard to hit. 8 ms with 100% reliability is.
8ms is incredibly slow for realtime systems. A difficult target for a control system would be three orders of magnitude lower than that on a full size OS like linux.
VxWorks is being partly phased out too, last I heard wind river was pushing people towards their own linux based offering. Speaking from experience it's absolutely miserable to develop on because of the licensing model and reliance on the Eclipse IDE. If you need a low level system FreeRTOS/CMSIS is a far better option and if you want linux compatibility (frequently just the convenience of ssh and rsync) just use real time linux.
I suppose the CPU on this machine may be severely underclocked because of the power requirements.
My application was simulcast for a software defined radio. It was running on a TMS320 DSP (192MHz) with DSP/BIOS as the operating system. Timing was driven by a GPS 1PPS pulse connected to an interrupt pin. The absolute response time was around 100ns and repeatability was of the order of 10 nanoseconds (dominated by the drift between the 1PPS and the processor's clock).
PowerPC is popular because there are a couple companies out there that are taking these parts and running them through radiation beams to certify them for spaceflight (essentially profiling their failure modes so devs can account for them). ARM isn't popular yet as they tend to make it difficult to license designs for lower volume silicon but RISC-V in particular is gaining traction here very rapidly.
If my impression is correct, this is based on Linux kernel.
Since other comments appearing not mentioning that, just add here in case someone mistaken it with a custom embedded OS. (Although I know little about embedded OS either).
HFTs are in the nanosecond range... so probably not that hard.
Also, my ping to remote servers is often 5ms...