31.9 Efficiency Considerations

The mutual exclusion mechanisms provided by Ice differ in their size and speed. Table 31.1 shows the relative sizes for Windows and Linux (both on Intel architec­tures).

Table 31.1. Size of mutual exclusion primitives.

Primitive	Size in Bytes (Windows)	Size in Bytes (Linux)
Mutex	24	24
RecMutex	28	28
RWRecMutex	172	184
Monitor<Mutex>	72	76
Monitor<RecMutex>	76	80

Table 31.2 shows the relative performance of the different synchronization primi­tives under conditions of no lock contention.

Table 31.2. Relative performance of mutual exclusion primitives.

Primitive	Speed (Windows)	Speed (Linux)
Mutex	1.00	1.00
RecMutex	1.02	1.25
RWRecMutex	12.24 (read lock) 22.45 (write lock)	3.23 (read lock) 3.40 (write lock)
Monitor<Mutex>	0.86	1.09
Monitor<RecMutex>	0.90	1.30

Note that the size and performance differences between a mutex and a recursive mutex are negligible, so you should use a non-recursive mutex only if you have a very small critical region that is accessed repeatedly in a tight loop. Similarly, monitors perform much the same as mutexes (but do have a size penalty).

Read-write locks incur a substantial cost in both size and speed (especially under Windows), so you should use a read-write lock only if you can truly exploit the additional parallelism afforded by multiple readers and if a substantial amount of work is performed inside the critical region.