UnixBench数据统计

Unixbench顾名思义是一款开源的Unix“综合测试跑分软件”,可以相对客观的评价一套Unix系统——主要是硬件平台的性能状况。本来打算用它跑几个新服务器的压力稳定测试,整理了一下跑分结果,就地说说它的测评项吧。

硬件是一台Dell的R320服务器,搭配Xeon E5-2430 CPU (1.8GHz * 4 core),  16G Ram, 板载的Sata RAID1, 使用的是CentOS6.4的最新版。要发的牢骚是,Centos6.0竟然不支持板载的BroadCom 双千网卡,只能去下载了最新的Centos。

System: localhost.localdomain: GNU/Linux
OS: GNU/Linux — 2.6.32-358.el6.x86_64 — #1 SMP Fri Feb 22 00:31:26 UTC 2013
Machine: x86_64: x86_64
Language: en_US.utf8 (charmap=”UTF-8″, collate=”UTF-8″)
CPUs: 0: Intel(R) Xeon(R) CPU E5-2403 0 @ 1.80GHz (3600.2 bogomips)
Hyper-Threading, x86-64, MMX, Physical Address Ext, SYSENTER/SYSEXIT, SYSCALL/SYSRET, Intel virtualization
1: Intel(R) Xeon(R) CPU E5-2403 0 @ 1.80GHz (3600.2 bogomips)
Hyper-Threading, x86-64, MMX, Physical Address Ext, SYSENTER/SYSEXIT, SYSCALL/SYSRET, Intel virtualization
2: Intel(R) Xeon(R) CPU E5-2403 0 @ 1.80GHz (3600.2 bogomips)
Hyper-Threading, x86-64, MMX, Physical Address Ext, SYSENTER/SYSEXIT, SYSCALL/SYSRET, Intel virtualization
3: Intel(R) Xeon(R) CPU E5-2403 0 @ 1.80GHz (3600.2 bogomips)
Hyper-Threading, x86-64, MMX, Physical Address Ext, SYSENTER/SYSEXIT, SYSCALL/SYSRET, Intel virtualization

由于是在跑Serve的操作系统,没有安装桌面,没有办法测试2D/3D等图形性能,在此直接略过。考虑到很多测试项目比较晦涩,只能根据字面意思,翻译一下:

  • Dhrystone 2 using register variables:dhrystone 2变量赋值,主要取决于整数性能
  • Double-Precision Whetstone:双精度Whetstone,浮点数为主的计算性能
  • Execl Throughpu:吞吐能力为主的测试
  • File Copy n bufsize M maxblocks: 不同情况下的文件拷贝能力数字越小,文件越琐碎
  • Pipe Throughput:管道吞吐,Unix特有的管道操作性能
  • Pipe-based Context Switching:基于管道的上下文切换性能
  • Process Creation:线程创建,主要是fork操作的性能
  • Shell Scripts(1 concurrent):1个shell实例的性能
  • System Call Overhead:系统调用消耗

得出了1个进程和4个进程的得分如下,我突发奇想的把1个进程的得分乘以4,然后跟4个进程的得分比较,得出了Rate项。

Test 1 process 4 processes Rate
Dhrystone 2 using register variables 19981976.8 79632697.3 99.63%
Double-Precision Whetstone 1183.1 4732.6 100.00%
Execl Throughput 2564.8 10337.1 100.76%
File Copy 1024 bufsize 2000 maxblocks 526601.2 592363.1 28.12%
File Copy 256 bufsize 500 maxblocks 140621.7 156159.0 27.76%
File Copy 4096 bufsize 8000 maxblocks 1275103.1 1856036.2 36.39%
Pipe Throughput 914092.2 3626326.4 99.18%
Pipe-based Context Switching 132192.1 790645.6 149.53%
Process Creation 9368.7 36353.8 97.01%
Shell Scripts (1 concurrent) 4491.6 13295.3 74.00%
Shell Scripts (8 concurrent) 1583.9 1787.9 28.22%
System Call Overhead 1393976.5 4323537.5 77.54%

貌似看不出什么东西吧,我又启动了一台虚拟机,配置2Core + 1GRam:

System: www.litrin.net: GNU/Linux
OS: GNU/Linux — 3.2.0-41-virtual — #66-Ubuntu SMP Thu Apr 25 03:47:17 UTC 2013
Machine: x86_64: x86_64
Language: en_US.utf8 (charmap=”UTF-8″, collate=”UTF-8″)
CPUs: 0: QEMU Virtual CPU version 1.0 (5188.1 bogomips)
x86-64, MMX, Physical Address Ext, SYSENTER/SYSEXIT, SYSCALL/SYSRET, Intel virtualization
1: QEMU Virtual CPU version 1.0 (5188.1 bogomips)
x86-64, MMX, Physical Address Ext, SYSENTER/SYSEXIT, SYSCALL/SYSRET, Intel virtualization
Uptime: 13:16:39 up 6 days, 1:18, 1 user, load average: 0.20, 0.09, 0.07; runlevel 2

由于只有双核,只能将Rate项的加权变成了2,得出了下表:

Dhrystone 2 using register variables 29937725.6 58766087.7 98.15%
Double-Precision Whetstone 3262.6 6515.3 99.85%
Execl Throughput 3428.8 6803.6 99.21%
File Copy 1024 bufsize 2000 maxblocks 788211.1 1036546.9 65.75%
File Copy 256 bufsize 500 maxblocks 247884.5 311322.5 62.80%
File Copy 4096 bufsize 8000 maxblocks 1513292.0 2352853.6 77.74%
Pipe Throughput 1904600.9 3652266.9 95.88%
Pipe-based Context Switching 237342.9 454277.1 95.70%
Process Creation 4487.3 17813.4 198.49%
Shell Scripts (1 concurrent) 5052.4 11868.5 117.45%
Shell Scripts (8 concurrent) 1547.1 1585.5 51.24%
System Call Overhead 3352978.0 5130982.8 76.51%

从数字上看,KVM方式的虚拟主机在多进程的表现上貌似比物理机好很多,但很明显的是,KVM最大拖后腿的地方在于IO。

大部分的得分跟bogomips有直接关系,而bogomips跟CPU频率挂钩,虚拟机的宿主主机是i5 2.6,相比Xeon  E5 1.8G是差了不少,但结果却反超,这不科学啊!

推荐阅读:
Faiss的多线程效率问题

过去的一周,被AI组的同事拉去 Read more

CPU各级缓存

被问起CPU的各级缓存,才想起 Read more

PMU Event counter的使用状况检测

题目用中文反而有点绕,How Read more

AMD Rome benchmark数据到架构特征推导

这几天,拿到了一套最新的AMD Read more

“UnixBench数据统计”的一个回复

  1. 4. Performance comparison using SPECfp*_rate_base2006 benchmark result at the same TDP. Baseline score of 271 on previous-generation 2S Intel® Xeon® processor X5690 based on best publication to http://www.spec.org using Intel® Compiler 12.1 as of January 17, 2012. For details, please see: http://www.spec.org/cpu2006/results/res2012q1/cpu2006-20111219-19195.html. New score of 466 based on Intel internal measured estimates using an Intel platform, code named Canoe Pass, with two Intel® Xeon® processor E5-2680, Intel® Turbo Boost Technology enabled, Enhanced Intel SpeedStep® Technology enabled, Intel® Hyper-Threading Technology (Intel® HT Technology) enabled, 64 GB RAM, Intel® Compiler 12.1, THP disabled, Red Hat Enterprise Linux* Server 6.1.

发表评论

电子邮件地址不会被公开。 必填项已用*标注

请补全下列算式: *

此站点使用Akismet来减少垃圾评论。了解我们如何处理您的评论数据