测试软件：sysbench-mutex

1、测试在大并发的请求mutex的时候，系统的性能
2、测试环境，虚拟机4线程，PC服务器96线程

1. root@pm1:/root# cat /proc/cpuinfo |grep "model name"
   
2. model name : Intel(R) Core(TM) i3-2100 CPU @ 3.10GHz
   
3. model name : Intel(R) Core(TM) i3-2100 CPU @ 3.10GHz
   
4. model name : Intel(R) Core(TM) i3-2100 CPU @ 3.10GHz
   
5. model name : Intel(R) Core(TM) i3-2100 CPU @ 3.10GHz

1. [root@ ~]# cat /proc/cpuinfo |grep "model name"
   
2. model name    : Intel(R) Xeon(R) CPU E7-4860 v2 @ 2.60GHz
   
3. model name    : Intel(R) Xeon(R) CPU E7-4860 v2 @ 2.60GHz
   
4. model name    : Intel(R) Xeon(R) CPU E7-4860 v2 @ 2.60GHz
   
5. model name    : Intel(R) Xeon(R) CPU E7-4860 v2 @ 2.60GHz
   
6. model name    : Intel(R) Xeon(R) CPU E7-4860 v2 @ 2.60GHz
   
7. ................................................

3、运行测试
虚拟机环境使用最大4线程

1. root@pm1:/root# sysbench --test=mutex --num-threads=4 --mutex-num=2000 --mutex-locks=10000 --mutex-loops=5000 run
   
2. sysbench 0.5: multi-threaded system evaluation benchmark
   
3. 
   
4. Running the test with following options:
   
5. Number of threads: 4
   
6. Random number generator seed is 0 and will be ignored
   
7. 
   
8. 
   
9. Threads
   
10. 
    
11. 
    
12. General statistics:
    
13.     total time: 0.0060s
    
14.     total number of events: 4
    
15.     total time taken by event execution: 0.0144s
    
16.     response time:
    
17.          min: 0.72ms
    
18.          avg: 3.61ms
    
19.          max: 5.75ms
    
20.          approx. 95 percentile: 5.75ms
    
21. 
    
22. Threads fairness:
    
23.     events (avg/stddev): 1.0000/0.00
    
24.     execution time (avg/stddev): 0.0036/0.00

PC服务器使用4线程

1. [root@ ~]# sysbench --test=mutex --num-threads=4 --mutex-num=2000 --mutex-locks=10000 --mutex-loops=5000 run
   
2. sysbench 0.5: multi-threaded system evaluation benchmark
   
3. 
   
4. Running the test with following options:
   
5. Number of threads: 4
   
6. Random number generator seed is 0 and will be ignored
   
7. 
   
8. 
   
9. Threads
   
10. 
    
11. 
    
12. General statistics:
    
13.     total time: 0.0210s
    
14.     total number of events: 4
    
15.     total time taken by event execution: 0.0819s
    
16.     response time:
    
17.          min: 19.42ms
    
18.          avg: 20.49ms
    
19.          max: 20.93ms
    
20.          approx. 95 percentile: 20.94ms
    
21. 
    
22. Threads fairness:
    
23.     events (avg/stddev): 1.0000/0.00
    
24.     execution time (avg/stddev): 0.0205/0.00

对比上面的结果，mutex跟线程数的多少没有关系，而是与CPU主频有关系
下面在看一组96线程的mutex竞争

1. [root@ ~]# sysbench --test=mutex --num-threads=96 --mutex-num=2000 --mutex-locks=10000 --mutex-loops=5000 run
   
2. sysbench 0.5: multi-threaded system evaluation benchmark
   
3. 
   
4. Running the test with following options:
   
5. Number of threads: 96
   
6. Random number generator seed is 0 and will be ignored
   
7. 
   
8. 
   
9. Threads
   
10. 
    
11. 
    
12. General statistics:
    
13.     total time: 0.5325s
    
14.     total number of events: 96
    
15.     total time taken by event execution: 48.1514s
    
16.     response time:
    
17.          min: 428.05ms
    
18.          avg: 501.58ms
    
19.          max: 532.18ms
    
20.          approx. 95 percentile: 530.29ms
    
21. 
    
22. Threads fairness:
    
23.     events (avg/stddev): 1.0000/0.00
    
24.     execution time (avg/stddev): 0.5016/0.02

用满线程数测试mutex反而更慢了，也就是说mutex的竞争与线程是没有关系的
在看下系统CPU信息，是4个物理CPU，12核，96线程的机器

1. [root@ ~]# grep "core id" /proc/cpuinfo | sort -u | wc -l
   
2. 12
   
3. [root@ ~]# grep "physical id" /proc/cpuinfo | sort -u | wc -l
   
4. 4
   
5. [root@ ~]# grep "processor" /proc/cpuinfo | sort -u | wc -l
   
6. 96

综上，mutex请求的性能与CPU主频及物理CPU个数有关

来自 “ ITPUB博客 ” ，链接：http://blog.itpub.net/12219480/viewspace-1749359/，如需转载，请注明出处，否则将追究法律责任。

转载于:http://blog.itpub.net/12219480/viewspace-1749359/