Locks - Measure Various Locking Strategies and their cost

The following table shows the results of timing various locking strategies with and without contention from competing threads. I also measure the cost to malloc memory with and without contention. The locking strategies are:

Windows	Linux
Interlock	Atomic instructions
Critical Section	Pthread Mutex

Counting in one thread, sleep(0) in 2nd thread.
Shared counting in two threads
Interlock (atomic instructions to set/test a variable) in one thread, sleep(0) in 2nd thread.
Interlock with two threads trying to set/test the same variable.
Pthread mutex (Linux) or Critical section (Windows) in single thread, sleep(0) in 2nd thread.
Mutex or Critical section with two theads sharing lock.
Memory allocation (malloc) in single thread, sleep(0) in 2nd thread.
Malloc occuring in two threads.

The following graph plots the relative performance of each run. The results are relative to test#1, which times how long it takes to increment a value in one thread while the 2nd thread was doing nothing.

/MD /O2 - MD is multithread DLL with level 2 optimization
/MD - MD is multithread DLL
/MT - MT is multithread static libraries.

Shared counting is more expensive than using an interlock (atomic) instruction. Shared counting is two threads touching the same memory using normal math operators. It is puzzling why the interlocks are both faster and thread safe. See article on Thread Locality Penalty
Interlocks or atomic instructions are fast both with and without contention compared to Mutex/Critical sections.
Mutex or Critical sections are expensive (as you would expect) and should be avoided.
Linux and Centos have faster 'malloc' than Windows, relative to time measured in test#1.
Ironically the Optimized windows code runs slower than the unoptimized.

The table is sorted in performance order (fastest to slowest) per Hardware configuration.
The table includes both Windows and Linux results.
* Click on column heading to sort. Supports multi-column sorts in order sorts applied.

Seconds	Test Description	Hardware
0.292112	Time interlock	Dell 670 2 dual/core 3 Ghz Linux 2.5? kernel
0.333655	Time couting in two threads	Dell 670 2 dual/core 3 Ghz Linux 2.5? kernel
0.365504	Time couting	Dell 670 2 dual/core 3 Ghz Linux 2.5? kernel
0.381036	Time interlock with collison	Dell 670 2 dual/core 3 Ghz Linux 2.5? kernel
0.503537	Time mutex/critical_section	Dell 670 2 dual/core 3 Ghz Linux 2.5? kernel
0.541135	Time malloc	Dell 670 2 dual/core 3 Ghz Linux 2.5? kernel
0.565152	Time malloc with contention	Dell 670 2 dual/core 3 Ghz Linux 2.5? kernel
0.588693	Time mutex/critical_section with collison	Dell 670 2 dual/core 3 Ghz Linux 2.5? kernel

0.154479	Time couting	Dell 690 Centos
0.167945	Time interlock	Dell 690 Centos
0.215153	Time interlock with collison	Dell 690 Centos
0.286555	Time malloc	Dell 690 Centos
0.308464	Time mutex/critical_section	Dell 690 Centos
0.327101	Time couting in two threads	Dell 690 Centos
0.452401	Time malloc with contention	Dell 690 Centos
0.844744	Time mutex/critical_section with collison	Dell 690 Centos

0.094684	Time couting	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi /O2)
0.107344	Time interlock	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi /O2)
0.124729	Time mutex/critical_section	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi /O2)
0.195459	Time interlock with collison	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi /O2)
0.212798	Time malloc	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi /O2)
0.259603	Time couting in two threads	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi /O2)
0.763054	Time mutex/critical_section with collison	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi /O2)
1.223511	Time malloc with contention	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi /O2)

0.109211	Time couting	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi)
0.123599	Time interlock	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi)
0.141559	Time mutex/critical_section	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi)
0.213573	Time interlock with collison	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi)
0.214477	Time malloc	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi)
0.276162	Time couting in two threads	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi)
0.961129	Time mutex/critical_section with collison	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi)
1.188000	Time malloc with contention	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /MD /Zi)

0.093182	Time couting	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /Mt /O2)
0.106709	Time interlock	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /Mt /O2)
0.156504	Time interlock with collison	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /Mt /O2)
0.159648	Time mutex/critical_section	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /Mt /O2)
0.203873	Time malloc	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /Mt /O2)
0.257813	Time couting in two threads	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /Mt /O2)
0.793430	Time mutex/critical_section with collison	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /Mt /O2)
1.268256	Time malloc with contention	HP 4 dual/core 2.8 Ghz Intel/Xeon Windows XP32 Professional (cl /D WIN32 /Mt /O2)

0.230114	Time couting	Windows XP - Laptop M1710
0.254910	Time interlock	Windows XP - Laptop M1710
0.286393	Time mutex/critical_section	Windows XP - Laptop M1710
0.338414	Time interlock with collison	Windows XP - Laptop M1710
0.443224	Time mutex/critical_section with collison	Windows XP - Laptop M1710
1.691142	Time malloc	Windows XP - Laptop M1710