Real-time scalability of nested spin locks

When a real-time system is realized on a shared-memory multiprocessor, the worst-case performance of interprocessor synchronizations is one of the most important issues. In this paper, scalability of the maximum execution times of critical sections guarded by nested spin locks is discussed. With the simplest method, the maximum execution times become O(n/sup m/), where n is the number of contending processors and m is the maximum nesting level of locks. In this paper, we propose an algorithm with which this order can be reduced to O(n/spl middot/e/sup m/) and demonstrate its effectiveness when m=2 through performance measurements.