3 ms·
The problem with a centralized system is that is a single point of failure. If the central system went out, things would get Very Bad, Very Fast.
by Turing_Machine 9y ago
The problem with a centralized system is that is a single point of failure. If the central system went out, things would get Very Bad, Very Fast.
- tyingq 9y agoThere's probably some version of a peer-to-peer "swarm" that gains some benefits though.
- Turing_Machine 9y agoYeah, there would be advantages in the cars exchanging data, but you could have that while still leaving the final decisions up to the individual car (rather than being dictated by a centralized controller). That's more or less what ants and bees do -- they exchange information with each other, but each unit is still quasi-independent.
- nealabq 9y agoIt doesn't have to be centralized. If you think of the cars as data packets and the roads as cat-6 cables, then you can see how redundancy and alternate routing can be built in.
- csours 9y agoYou'd want that protocol to be EXTREMELY collision resistant.
- Turing_Machine 9y agoRight. You don't get an automatic retry with cars.
- stevekemp 9y agoIndeed, and you would also want to avoid fragmentation of packets (cars).
- barrkel 9y agoI'd like guaranteed once delivery, personally. While I take your point, many of the system behaviours of IP networking rely on inferences from failure. Information gathering needs to be a bit more proactive and connected without it. Eg determining a jam shouldn't be a matter of noticing that cars sent through a particular area are getting stuck ("dropped") but something that can estimate viable throughput before triggering phase changes like jams.
- CoryG89 9y agoTotally agree. This is not something I would have considered as a child, but it seems clear that some decentralization is needed. I think the proper solution is definitely some kind of p2p protocol with cars exchanging information.