Dining philosophers that tolerate malicious crashes

We present a solution to the problem of dining philosophers. Our solution tolerates malicious crashes. In a malicious crash the failed process behaves arbitrarily for a finite time and then ceases all operation undetectably to other processes. The tolerance of our solution is achieved by the combination of stabilization and crash failure locality. Stabilization allows our program to recover from an arbitrary state. Crash failure locality ensures that only a limited number of processes are affected by a process crash. The crash failure locality of our solution is optimal. Finally, we argue that the malicious crash fault model and its extensions are worthy of further study as they admit tolerances that are not achieved under stronger fault models and are unnecessary under weaker fault models.