首先需求了解基本的信息
1 宿主機CPU特性查看
使用virsh nodeinfo可以看到一些基本信息
virsh nodeinfo
CPU model: x86_64
CPU(s): 32
CPU frequency: 1200 MHz
CPU socket(s): 1
Core(s) per socket: 8
Thread(s) per core: 2
NUMA cell(s): 2
Memory size: 132119080 KiB
使用virsh capabilities可以查看物理機CPU的詳細信息，包括物理CPU個數(shù)，每個CPU的核數(shù)，是否開了超線程。
/p>
p>virsh capabilities
capabilities>
host>
uuid>36353332-3030-3643-5534-3235445a564a/uuid>
cpu>
arch>x86_64/arch>
model>SandyBridge/model>
vendor>Intel/vendor>
topology sockets='1' cores='8' threads='2'/>
feature name='erms'/>
feature name='smep'/>
...
/cpu>
power_management>
suspend_disk/>
/power_management>
migration_features>
live/>
uri_transports>
uri_transport>tcp/uri_transport>
/uri_transports>
/migration_features>
topology>
cells num='2'>
cell id='0'>
cpus num='16'>
cpu id='0' socket_id='0' core_id='0' siblings='0,16'/>
...
cpu id='23' socket_id='0' core_id='7' siblings='7,23'/>
/cpus>
/cell>
cell id='1'>
cpus num='16'>
cpu id='8' socket_id='1' core_id='0' siblings='8,24'/>
...
cpu id='31' socket_id='1' core_id='7' siblings='15,31'/>
/cpus>
/cell>
/cells>
/topology>
secmodel>
model>none/model>
doi>0/doi>
/secmodel>
secmodel>
model>dac/model>
doi>0/doi>
/secmodel>
/host>
...
/capabilities>
使用virsh freecell命令查看可以當前空閑內(nèi)存
virsh freecell --all
0: 787288 KiB
1: 94192 KiB
--------------------
Total: 881480 KiB
物理CPU的特性也可以通過/proc/cpuinfo查看
cat /proc/cpuinfo
rocessor : 0
vendor_id : GenuineIntel
cpu family : 6
model : 62
model name : Intel(R) Xeon(R) CPU E5-2640 v2 @ 2.00GHz
stepping : 4
cpu MHz : 1200.000
cache size : 20480 KB
physical id : 0
siblings : 16
core id : 0
cpu cores : 8
apicid : 0
initial apicid : 0
fpu : yes
fpu_exception : yes
cpuid level : 13
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc arch_perfmon pebs bts rep_good xtopology nonstop_tsc aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm ida arat epb xsaveopt pln pts dts tpr_shadow vnmi flexpriority ept vpid fsgsbase smep erms
bogomips : 3990.67
clflush size : 64
cache_alignment : 64
address sizes : 46 bits physical, 48 bits virtual
power management:
...
綜合上面的信息，我們可以得出以下信息：
1) 物理CPU為 E5-2640V2，為8核2顆，開啟了超線程，在物理機系統(tǒng)上可以看到32個CPU；
2) 物理機內(nèi)存為128G
2 虛擬機CPU使用情況查看
可以使用virsh vcpuinfo命令查看虛擬機vcpu和物理CPU的對應關系
virsh vcpuinfo 21
VCPU: 0
CPU: 25
State: running
CPU time: 10393.0s
CPU Affinity: --------yyyyyyyy--------yyyyyyyy
VCPU: 1
CPU: 8
State: running
CPU time: 7221.2s
CPU Affinity: --------yyyyyyyy--------yyyyyyyy
...
可以看到vcpu0被調(diào)度到物理機CPU25上，目前是使用狀態(tài)，使用時間是10393.0s
CPU Affinity: --------yyyyyyyy--------yyyyyyyy
yyyyyyy表示可以使用的物理CPU內(nèi)部的邏輯核，可以看到這臺虛擬機可以在8-15， 24-31這些cpu之間調(diào)度，為什么不能使用0-7，16-23這些CPU呢，是因為系統(tǒng)的自動numa平衡服務在發(fā)生作用，一個虛擬機默認只能使用同一顆物理CPU內(nèi)部的邏輯核。
使用emulatorpin可以查看虛擬機可以使用那些物理邏輯CPU
virsh # emulatorpin 21
emulator: CPU Affinity
----------------------------------
*: 0-31
可以看到0-31我們都可以使用，意味這我們也可以強制將CPU調(diào)度到任何CPU上。
3 在線pinning虛擬機的cpu
強制讓虛擬機只能在26-31這些cpu之間調(diào)度
virsh emulatorpin 21 26-31 --live
查看結(jié)果
virsh emulatorpin 21
emulator: CPU Affinity
----------------------------------
*: 26-31
查看vcpu info
virsh vcpuinfo 21
VCPU: 0
CPU: 28
State: running
CPU time: 10510.5s
CPU Affinity: --------------------------yyyyyy
VCPU: 1
CPU: 28
State: running
CPU time: 7289.7s
CPU Affinity: --------------------------yyyyyy
.../p>
p>
查看xml文件
virsh # dumpxml 21
domain type='kvm' id='21'>
name>cacti-230/name>
uuid>23a6455c-5cd1-20cd-ecfe-2ba89be72c41/uuid>
memory unit='KiB'>4194304/memory>
currentMemory unit='KiB'>4194304/currentMemory>
vcpu placement='static'>4/vcpu>
cputune>
emulatorpin cpuset='26-31'/>
/cputune>
我們也可以強制vcpu和物理機cpu一對一的綁定
強制vcpu 0和物理機cpu 28綁定
強制vcpu 1和物理機cpu 29綁定
強制vcpu 2和物理機cpu 30綁定
強制vcpu 3和物理機cpu 31綁定
virsh vcpupin 21 0 28
virsh vcpupin 21 1 29
virsh vcpupin 21 2 30
virsh vcpupin 21 3 31
查看xml文件，生效了
virsh # dumpxml 21
domain type='kvm' id='21'>
name>cacti-230/name>
uuid>23a6455c-5cd1-20cd-ecfe-2ba89be72c41/uuid>
memory unit='KiB'>4194304/memory>
currentMemory unit='KiB'>4194304/currentMemory>
vcpu placement='static'>4/vcpu>
cputune>
vcpupin vcpu='0' cpuset='28'/>
vcpupin vcpu='1' cpuset='29'/>
vcpupin vcpu='2' cpuset='30'/>
vcpupin vcpu='3' cpuset='31'/>
emulatorpin cpuset='26-31'/>
/cputune>
是vcpuino命令查看，可以看到配置生效了
virsh vcpuinfo 22
VCPU: 0
CPU: 28
State: running
CPU time: 1.8s
CPU Affinity: ----------------------------y---
VCPU: 1
CPU: 29
State: running
CPU time: 0.0s
CPU Affinity: -----------------------------y--
...
4 cpu pinning簡單的性能測試
cpu pinning到底對cpu的性能影響有多大，進行了一個簡單的測試。
測試環(huán)境
硬件
ntel(R) Xeon(R) CPU X5650 @ 2.67GHz 2顆
軟件
centos 7 update到內(nèi)核 3.10.0-123.8.1.el7.x86_64
虛擬機
centos 7 update到內(nèi)核 3.10.0-123.8.1.el7.x86_64
虛擬機 cpu 1顆
測試工具
unixbench 5.1.2
測試結(jié)果
不做cpu綁定
1 CPU in system; running 1 parallel copy of tests
Dhrystone 2 using register variables 28890881.0 lps (10.0 s, 7 samples)
Double-Precision Whetstone 3880.4 MWIPS (9.0 s, 7 samples)
Execl Throughput 4146.3 lps (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks 1051084.3 KBps (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks 286552.2 KBps (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks 2142638.4 KBps (30.0 s, 2 samples)
Pipe Throughput 1726807.0 lps (10.0 s, 7 samples)
Pipe-based Context Switching 322865.5 lps (10.0 s, 7 samples)
Process Creation 13662.4 lps (30.0 s, 2 samples)
Shell Scripts (1 concurrent) 5955.4 lpm (60.0 s, 2 samples)
Shell Scripts (8 concurrent) 713.1 lpm (60.0 s, 2 samples)
System Call Overhead 2138318.1 lps (10.0 s, 7 samples)
System Benchmarks Index Values BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 28890881.0 2475.7
Double-Precision Whetstone 55.0 3880.4 705.5
Execl Throughput 43.0 4146.3 964.2
File Copy 1024 bufsize 2000 maxblocks 3960.0 1051084.3 2654.3
File Copy 256 bufsize 500 maxblocks 1655.0 286552.2 1731.4
File Copy 4096 bufsize 8000 maxblocks 5800.0 2142638.4 3694.2
Pipe Throughput 12440.0 1726807.0 1388.1
Pipe-based Context Switching 4000.0 322865.5 807.2
Process Creation 126.0 13662.4 1084.3
Shell Scripts (1 concurrent) 42.4 5955.4 1404.6
Shell Scripts (8 concurrent) 6.0 713.1 1188.4
System Call Overhead 15000.0 2138318.1 1425.5
========
System Benchmarks Index Score 1444.7
做了cpu綁定
1 CPU in system; running 1 parallel copy of tests
Dhrystone 2 using register variables 29812559.6 lps (10.0 s, 7 samples)
Double-Precision Whetstone 3928.7 MWIPS (8.9 s, 7 samples)
Execl Throughput 4314.4 lps (30.0 s, 2 samples)
File Copy 1024 bufsize 2000 maxblocks 1068627.9 KBps (30.0 s, 2 samples)
File Copy 256 bufsize 500 maxblocks 291834.2 KBps (30.0 s, 2 samples)
File Copy 4096 bufsize 8000 maxblocks 2052612.8 KBps (30.0 s, 2 samples)
Pipe Throughput 1737466.2 lps (10.0 s, 7 samples)
Pipe-based Context Switching 326839.9 lps (10.0 s, 7 samples)
Process Creation 14234.5 lps (30.0 s, 2 samples)
Shell Scripts (1 concurrent) 6040.8 lpm (60.0 s, 2 samples)
Shell Scripts (8 concurrent) 717.4 lpm (60.1 s, 2 samples)
System Call Overhead 2149194.4 lps (10.0 s, 7 samples)
System Benchmarks Index Values BASELINE RESULT INDEX
Dhrystone 2 using register variables 116700.0 29812559.6 2554.6
Double-Precision Whetstone 55.0 3928.7 714.3
Execl Throughput 43.0 4314.4 1003.4
File Copy 1024 bufsize 2000 maxblocks 3960.0 1068627.9 2698.6
File Copy 256 bufsize 500 maxblocks 1655.0 291834.2 1763.3
File Copy 4096 bufsize 8000 maxblocks 5800.0 2052612.8 3539.0
Pipe Throughput 12440.0 1737466.2 1396.7
Pipe-based Context Switching 4000.0 326839.9 817.1
Process Creation 126.0 14234.5 1129.7
Shell Scripts (1 concurrent) 42.4 6040.8 1424.7
Shell Scripts (8 concurrent) 6.0 717.4 1195.7
System Call Overhead 15000.0 2149194.4 1432.8
========
System Benchmarks Index Score 1464.1
比較
綜合得分
綁定 1464.1 不綁定 1444.7
綜合得分 性能提升 1.34%
浮點運算
綁定 3928.7 不綁定 3880.4
浮點運算 性能提升 1.24%
