本文包含以下内容:
宿主机CPU特性查看
虚拟机CPU特性查看
KVM虚拟机cpu pinning如何配置
cpu pinning的应用场景
cpu pinning简单的性能测试
首先需求了解基本的信息
1 宿主机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可以查看物理机CPU的详细信息,包括物理CPU个数,每个CPU的核数,是否开了超线程。
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命令查看可以当前空闲内存
virsh freecell --all 0: 787288 KiB 1: 94192 KiB -------------------- Total: 881480 KiB
物理CPU的特性也可以通过/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: ...
综合上面的信息,我们可以得出以下信息:
1) 物理CPU为 E5-2640V2,为8核2颗,开启了超线程,在物理机系统上可以看到32个CPU;
2) 物理机内存为128G
2 虚拟机CPU使用情况查看
可以使用virsh vcpuinfo命令查看虚拟机vcpu和物理CPU的对应关系
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被调度到物理机CPU25上,目前是使用状态,使用时间是10393.0s
CPU Affinity: --------yyyyyyyy--------yyyyyyyy
yyyyyyy表示可以使用的物理CPU内部的逻辑核,可以看到这台虚拟机可以在8-15, 24-31这些cpu之间调度,为什么不能使用0-7,16-23这些CPU呢,是因为系统的自动numa平衡服务在发生作用,一个虚拟机默认只能使用同一颗物理CPU内部的逻辑核。
使用emulatorpin可以查看虚拟机可以使用那些物理逻辑CPU
virsh # emulatorpin 21 emulator: CPU Affinity ---------------------------------- *: 0-31
可以看到0-31我们都可以使用,意味这我们也可以强制将CPU调度到任何CPU上。
3 在线pinning虚拟机的cpu
强制让虚拟机只能在26-31这些cpu之间调度
virsh emulatorpin 21 26-31 --live
查看结果
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 ...
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>
我们也可以强制vcpu和物理机cpu一对一的绑定
强制vcpu 0和物理机cpu 28绑定
强制vcpu 1和物理机cpu 29绑定
强制vcpu 2和物理机cpu 30绑定
强制vcpu 3和物理机cpu 31绑定
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技术应用场景
pinning前后效果比较
没有做cpu pinning
前面的cpu压力和后面的cpu压力相差很大
做了cpu pinning
前后CPU性能比较平衡了
cpupin实际libvirt也是通过cgroup来实现的,通过cgroup直接去限制kvm虚拟机进程也可以,以后分享给大家。
5 cpu pinning简单的性能测试
cpu pinning到底对cpu的性能影响有多大,进行了一个简单的测试。
测试环境
硬件
ntel(R) Xeon(R) CPU X5650 @ 2.67GHz 2颗
软件
centos 7 update到内核 3.10.0-123.8.1.el7.x86_64
虚拟机
centos 7 update到内核 3.10.0-123.8.1.el7.x86_64
虚拟机 cpu 1颗
测试工具
unixbench 5.1.2
测试结果
不做cpu绑定
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绑定
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
比较
综合得分
绑定 1464.1 不绑定 1444.7
综合得分 性能提升 1.34%
浮点运算
绑定 3928.7 不绑定 3880.4
浮点运算 性能提升 1.24%