Comment by dlinder
3 days ago
Token-passing networks were interesting for hard real time system analysis, as they featured deterministic access control instead of collision detection. They made possible extending Rate Monotonic Scheduling theory to the network layer as well. The priority work (priority ceiling protocol, multi level token priority schemes) survive today in token bucket rate limiting.
Interestingly, the CAN bus features deterministic access control through collision detection, using a bus topology rather than a ring topology. I've had bad experiences with CAN, but I believe they're due to hardware bugs and/or driver bugs rather than the underlying design.
Arbitration in CAN is still one of the coolest things I've seen at the hardware layer.
Being able to handle a collision without needing the higher priority message to skip a single beat seemed like magic to me when I first learned about it
the token bus combines the one-user-at-a time of a bus with the potential-token-loss of a ring.
IEEE 802.4 token bus standard was withdrawn
polled buses with a master are pretty common when you don't mind the performance hit and central-point-of-failure. i believe cable networks work this way.
Interesting, I'd never heard of token-bus networks, or IEEE 802.4, which was indeed a token-bus standard.
CAN isn't a polled bus, doesn't have a master, and doesn't impose a performance hit. However, on a bus network, every device on the bus is a single point of failure: if it starts jabbering, the network stops working.
everything can be deterministic until you lose the token.