several common synchronization mechanisms for a million iterations of a producer/consumer loop.
The three mechanisms tried are:
-
A pipe. Producer writes a byte after each element is produced. Consumer reads a byte before consuming element.
-
A pthread_cond_t condition variable.
-
A sem_t couting semaphore.
The output on my machine (manjaro linux) is:
evaitl@bb sync_test]$ ls
LICENSE Makefile README.md sync.c
[evaitl@bb sync_test]$ make
gcc -Wall -O2 -c -o sync.o sync.c
gcc -Wall -O2 -o sync sync.o -lpthread
[evaitl@bb sync_test]$ ./sync
Pipe. Q = 0, max_q_size 5956, underflows 0
Time: 0 S:695100442 nS
Cond var. Q = 0, max_q_size 104162, underflows 0
Time: 0 S:465343564 nS
Counting sem. Q = 0, max_q_size 19323, underflows 0
Time: 0 S:430964583 nS
[evaitl@bb sync_test]$ ./sync
Pipe. Q = 0, max_q_size 5299, underflows 0
Time: 0 S:739420086 nS
Cond var. Q = 0, max_q_size 89481, underflows 0
Time: 0 S:423448141 nS
Counting sem. Q = 0, max_q_size 2531, underflows 0
Time: 0 S:418388530 nS
[evaitl@bb sync_test]$ ./sync
Pipe. Q = 0, max_q_size 7543, underflows 0
Time: 0 S:694078542 nS
Cond var. Q = 0, max_q_size 334265, underflows 0
Time: 0 S:398508465 nS
Counting sem. Q = 0, max_q_size 56804, underflows 0
Time: 0 S:381196771 nS
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent