StackThreads/MP: version 0.77 User's Guide

11.11: Why both spin-locks and mutex?

Q: StackThreads/MP provides both spin-lock primitives and mutex locks. Why? And when should I use which?

A: Because both are useful on different situations. Mutex is useful when the critical section is not very short. When threads are contending on a mutex, these threads enter the waiting queue of the mutex and the workers schedule other runnable threads. The effect is to serialize computation within these critical sections and perform other tasks in parallel.

To achieve scalable speedup, you must be careful so that the total time spent on any critical section is far smaller than the ideal execution time on whatever number of processors you wish to scale. For example, if the program spends T seconds on a single processor and the total amount of time that must be spent in a critical section is X, you must make sure that T/P << X, where P is the maximum number of processors that you wish to run the program. Note that thread blocking takes a fairly long time, so the above X must account for blocking/unblocking threads on a mutex.

A common practice is to do your best to make critical sections as short as possible. A typical critical section is as short as 10 instructions. Spin-locks are useful for such short critical sections.