咋一个CPU使用100%，其他的还那么闲，怎么内核不会调度吗？

jasonsungblog

top - 10:10:24 up 35 min,  2 users,  load average: 6.63, 5.88, 4.53
Tasks: 314 total,   6 running, 308 sleeping,   0 stopped,   0 zombie
Cpu0  :  2.0%us, 16.8%sy,  0.0%ni, 80.9%id,  0.0%wa,  0.0%hi,  0.3%si,  0.0%st
Cpu1  :  1.0%us, 16.7%sy,  0.0%ni, 81.9%id,  0.0%wa,  0.0%hi,  0.3%si,  0.0%st
Cpu2  :  2.4%us, 25.4%sy,  0.0%ni, 71.9%id,  0.0%wa,  0.0%hi,  0.3%si,  0.0%st
Cpu3  :  0.0%us, 73.2%sy,  0.0%ni,  0.0%id,  0.0%wa,  0.0%hi, 26.8%si,  0.0%st
Mem:   3334980k total,  3202088k used,   132892k free,    12832k buffers
Swap:        0k total,        0k used,        0k free,   120920k cache

都是一些内核线程使用，软中断只发生在一个CPU，一个CPU用完了，其他CPU还很闲，怎么搞的？

crazyhadoop

这个可以调节网卡的中断配置来均衡网卡的中断。可以参考这个教程
http://www.mjmwired.net/kernel/Documentation/IRQ-affinity.txt

1. 
   
2. 1        ChangeLog:
   
3. 2                Started by Ingo Molnar <mingo@redhat.com>
   
4. 3                Update by Max Krasnyansky <maxk@qualcomm.com>
   
5. 4       
   
6. 5        SMP IRQ affinity
   
7. 6       
   
8. 7        /proc/irq/IRQ#/smp_affinity and /proc/irq/IRQ#/smp_affinity_list specify
   
9. 8        which target CPUs are permitted for a given IRQ source.  It's a bitmask
   
10. 9        (smp_affinity) or cpu list (smp_affinity_list) of allowed CPUs.  It's not
    
11. 10        allowed to turn off all CPUs, and if an IRQ controller does not support
    
12. 11        IRQ affinity then the value will not change from the default of all cpus.
    
13. 12       
    
14. 13        /proc/irq/default_smp_affinity specifies default affinity mask that applies
    
15. 14        to all non-active IRQs. Once IRQ is allocated/activated its affinity bitmask
    
16. 15        will be set to the default mask. It can then be changed as described above.
    
17. 16        Default mask is 0xffffffff.
    
18. 17       
    
19. 18        Here is an example of restricting IRQ44 (eth1) to CPU0-3 then restricting
    
20. 19        it to CPU4-7 (this is an 8-CPU SMP box):
    
21. 20       
    
22. 21        [root@moon 44]# cd /proc/irq/44
    
23. 22        [root@moon 44]# cat smp_affinity
    
24. 23        ffffffff
    
25. 24       
    
26. 25        [root@moon 44]# echo 0f > smp_affinity
    
27. 26        [root@moon 44]# cat smp_affinity
    
28. 27        0000000f
    
29. 28        [root@moon 44]# ping -f h
    
30. 29        PING hell (195.4.7.3): 56 data bytes
    
31. 30        ...
    
32. 31        --- hell ping statistics ---
    
33. 32        6029 packets transmitted, 6027 packets received, 0% packet loss
    
34. 33        round-trip min/avg/max = 0.1/0.1/0.4 ms
    
35. 34        [root@moon 44]# cat /proc/interrupts | grep 'CPU\|44:'
    
36. 35                   CPU0       CPU1       CPU2       CPU3      CPU4       CPU5        CPU6       CPU7
    
37. 36         44:       1068       1785       1785       1783         0          0           0         0    IO-APIC-level  eth1
    
38. 37       
    
39. 38        As can be seen from the line above IRQ44 was delivered only to the first four
    
40. 39        processors (0-3).
    
41. 40        Now lets restrict that IRQ to CPU(4-7).
    
42. 41       
    
43. 42        [root@moon 44]# echo f0 > smp_affinity
    
44. 43        [root@moon 44]# cat smp_affinity
    
45. 44        000000f0
    
46. 45        [root@moon 44]# ping -f h
    
47. 46        PING hell (195.4.7.3): 56 data bytes
    
48. 47        ..
    
49. 48        --- hell ping statistics ---
    
50. 49        2779 packets transmitted, 2777 packets received, 0% packet loss
    
51. 50        round-trip min/avg/max = 0.1/0.5/585.4 ms
    
52. 51        [root@moon 44]# cat /proc/interrupts |  'CPU\|44:'
    
53. 52                   CPU0       CPU1       CPU2       CPU3      CPU4       CPU5        CPU6       CPU7
    
54. 53         44:       1068       1785       1785       1783      1784       1069        1070       1069   IO-APIC-level  eth1
    
55. 54       
    
56. 55        This time around IRQ44 was delivered only to the last four processors.
    
57. 56        i.e counters for the CPU0-3 did not change.
    
58. 57       
    
59. 58        Here is an example of limiting that same irq (44) to cpus 1024 to 1031:
    
60. 59       
    
61. 60        [root@moon 44]# echo 1024-1031 > smp_affinity_list
    
62. 61        [root@moon 44]# cat smp_affinity_list
    
63. 62        1024-1031
    
64. 63       
    
65. 64        Note that to do this with a bitmask would require 32 bitmasks of zero
    
66. 65        to follow the pertinent one.

复制代码

jasonsungblog

回复 2# crazyhadoop
看了这个没有问题，这个好像是调节硬件中断的吧？看几个CPU分配的挺平均的。不知道为什么软中断只在一个CPU发生？