“A Read-Copy Update based parallel server for distributed crowd simulations” (paywalled) recently appeared in Journal of Supercomputing. Vigueras et al. apply RCU to crowd simulation (as you might guess from the title), which combines throughput (“faster is better”) and response-time requirements (“never trust any result older than 250 milliseconds”). Use of RCU instead of straight locking allows them to support crowds with three times the number of agents while still meeting their response-time requirements.
It is good to see academics experimenting with RCU. Even better, I learned of this only after seeing the paper. 😉
paulmckrcu 