- The physical machine used in this test is the editor's Dell n5010 notebook. Let's take a look at the configuration of the physical machine:
- Test Method
- Test report
- Cinebench r11.5 rendering Test
- Sisoftware Sandra 2013 lite Test
- Encryption/Decryption Performance
- Multi-core Efficiency Test
- Memory performance test
- Next, let's take a look at the disk I/O performance.
- Sisoftware Sander hard disk test
- 3dmark Test
- Pcmark Test
Reprinted from http://www.dadclab.com/archives/2554
I have read the introduction of the first virtual machine software for virtualization technology and evaluation, and the introduction of the second virtual machine software. The audience estimates that the Win8 + hyper-V virtual machine has some knowledge, today I want to share with you the test process, so that you can clearly understand the various data parameters in the evaluation process ......
The physical machine used in this test is the editor's Dell n5010 notebook. Let's take a look at the configuration of the physical machine:
Windows 7x64 SP1 flagship Edition
Linux testing environment:
Fedora 17 (kernel version: Linux 3.3.4-5)
Windows test environment:
Windows 7x64 SP1 ultimate
Windows 8x64 Professional
Because the virtual machines used for testing have different expressions on configurations, to ensure the accuracy and fairness of testing. Set the virtual machine to the following Configuration:
The CPU is a four-core or dual-core four-Thread
Memory is set to 2048 MB (2 GB)
Hard disk capacity is 30 GB
You can set the video memory to 128 MB and enable 2D/3D acceleration.
Remove unnecessary devices such as the drive and printer.
During the test, KVM and hyper-V cannot set the memory size and 2D/3D acceleration.
Windows 7x86 SP1 ultimate is installed in the virtual machine.
Operating System Images use the msdn version to ensure the system is clean.
Virtual Machine Software:
VMware Workstation v9.0.0 build-812388
VMware Tools for Windows 9.2.0
Test Software Version:
Pcmark 7 v1.04.0
3 dmark 06 v1.1.0
Super PI mod 1.5
Sisoftware Sandra 2013 Lite
Use the testing software mentioned in the previous section to test the Virtual Machine System.
The test data sample is presented as a table or a bar chart. It can be used for horizontal comparison.
In the general test, we use WinRAR, super Pi 1 m, and wprime 32 m.
This test focuses on testing the CPU performance.
The unit of super PI and wprime test results is seconds. The shorter the value, the better.
The larger the WinRAR test result, the better.
The data is as follows:
The bar chart is as follows:
From the above test results, we can see that the CPU performance of each virtual machine does not suffer much loss under the same operating system.
Cinebench r11.5 rendering Test
The test in this section is to check whether various virtual machines can perform large-scale rendering on the CPU, and whether there is a large performance deficiency in the CPU floating-point mode.
The larger the CPU test data, the better.
The larger the OpenGL test data result, the better. If no test data exists, the corresponding test fails.
From the above data, we can see that the CPU performance of the 17 ora 17 + KVM group is the lowest, only a long rendering time, the actual application is still possible.
The highest score is that the physical machine runs Windows 7. The score of the fedora 17 + KVM group is only 57% of the physical machine performance.
As to why KVM has suffered such a huge decline, the editor guessed that KVM has adjusted its use of CPU resources for the physical machine to ensure the performance of the physical machine.
The performance of other virtual machines is slightly different from that of physical machines.
In contrast, the OpenGL test is not supported by Virtual Machine graphics cards, so hpyer-V, KVM, and virtualbox cannot be tested.
The performance of vmware9 is slightly better, but it is not smooth during the animation playing process. It is a little better than that of the paint video. The performance of vmware9 is reduced by 55% and 44% compared with that of the graphics card of the physical machine.
Sisoftware Sandra 2013 lite Test
This step involves many tests.
Arithmetic processor Test
During the arithmetic processor test, Sandra 2013 will run the integer and logical operation benchmark program and floating point operation benchmark program for the CPU's Alu (arithmetic logical unit) and FPU (floating point unit, to test the processor's logical operation and floating point operation capabilities.
P.s. Unit explanation
Billions of GOPs operations per second
Billions of gips commands per second
Gflops billions of floating point operations per second
The larger the three units, the better.
From the above data:
In terms of the master machine performance, the performance of virtual machines that host the operating system with Win8 as the physical machine has been improved, and performance indicators have been improved by about 3%.
The performance of Win8 + hyper-V has reached 3.034 billion operations per second.
In the dhrystone integer ALU test, Win8 + vmware9 is the closest to the real performance of the physical machine in the virtual machine environment, reaching 33.78 gips (executing 3.378 billion commands per second ), the performance gap with physical machines is Only 0.14%.
The combination of Win8 + hyper-V is very close to the real performance of the physical machine, reaching 33.75 gips (executing 3.375 billion commands per second). The performance gap between Win8 + hyper-V and the physical machine is 0.23%.
Next, let's take a look at the floating point performance.
It is also the best performance of Win8 + hyper-V, reaching 2.727 billion floating point operations per second, which is about 6% higher than the physical machine equipped with Windows 7.
Here is an interesting situation where the physical machine score lags behind that of Win8 + hyper-v.
When Microsoft officially launched Win8, it said that Win8 is better than win7, so this result should be caused by Win8 performance better than win7.
Multimedia processor Testing
The multimedia processor test mainly tests the CPU's ability to process multimedia instructions and data. This test calls MMX, SSE, avx, and other instruction sets to generate a fragment map, in addition, each pixel in the figure is calculated in 255 cycles. The specific test item also includes two aspects: integer and floating point.
P.s. Unit explanation
Mpixel/S million pixels/second
The larger the value, the better.
In the multimedia processor test, the testing results of virtual machine systems with Win8 as the physical machine operating system are very good.
But the Virtual Machine of Fedora 17 + KVM is still quite tragic.
The n5010 + win7 test results are still good, reaching 48.12 million pixels/second
In this step, the CPU uses the encryption algorithm to perform operations and evaluate the computing capability of the CPU.
P.s. Unit explanation
Multi-core Efficiency Test
The multi-core efficiency test mainly tests the multi-core CPU efficiency and internal interconnection efficiency, it mainly uses data blocks of different sizes and chain lengths to test the core's ability to process data blocks and transmit data to another core for processing.
In this test, the more cores, the better the L2 or L3 processor will get in this test.
Memory performance test
This step mainly tests the memory performance, which reflects the running speed to a certain extent.
P.s. Unit explanation
GB/S: 1 Gbit/s. The larger the value, the better.
NS: the smaller the value, the better.
The memory performance is generally close to the physical machine level. Generally, applications on virtual machines are free of pressure ~
Next, let's take a look at the disk I/O performance.
Disk I/O testing, this section uses the physical hard disk test provided by sisoftware Sandra (only test read performance, write performance needs to erase hard disk data), and file system testing.
Use crystaldiskmark 3.0.1 to test hard disk reading. The depth is 5 and the data sample size is 1000 MB.
The test items in the table above test the random read/write capability and continuous read/write capability of the hard disk.
N5010 physical machine hard disk is a st320g 7200 RPM. The speed is even better.
If the score is low, the test drive letter is related to the system disk.
Sisoftware Sander hard disk test
Because many graphics cards in this part of the test do not support DX9, there are few VMS involved in the test. Data is for reference only.
This test is also one of the data samples for reference only. When Win8 + hyper-V tests the video playback performance, an error is reported directly. Windows 7/8 + virtualbox cannot start the test process.
For comparison, the editor adds the last bar chart !!
This time, the hydrology may have been the longest written by the editor, but it remains quite watery.
You can read a large amount of data from the above, and the full virtualization performance is really good, but in the actual test process of xiaobian, the full virtualization virtual machine occupies a lot of space when using system resources, the physical machine is basically in the suspended state.
Although hyper-V and KVM are much better, the official name of hyper-V is the parent partition operating system, given that after hyper-V is enabled, it will host the system (which is called another host system, which is not accurate. The host system is also equivalent to being "virtualized", so hyper-V may not be able to perform well in some projects.
The same is true for KVM. Since the small series did not carefully understand the working mode of KVM and the corresponding technical architecture, the time was too short to understand why KVM has such a test result.
As a veteran of virtualization software, VMWare has not disappointed its test scores and practical applications. The only depressing thing is the huge volume of VMware, compared with hyper-V, KVM, virtualbox, the "slim" virtualization software, VMWare is indeed a "Fat Man.