Virtualization tools analysis mapped into RING 0
Josef Horalek, Ondrej Marik, Sona Neradova, Stanislav Zitta
University of Pardubice, Faculty of Electrical Engineering and Informatics, Pardubice, Czech Republic [email protected], [email protected], [email protected], [email protected] Abstract— Article introduces the topic of virtualization and
testing methods of virtualization tools that use type 1 hypervisor. Individual tests and their evaluation, based on which using individual tools is recommended, are introduced.
I. INTRODUCTION
Nowadays, utilizing virtualization belongs to standard solutions companies and educational institutions recourse to more and more frequently. Virtualization provides many advantages, as discussed by [1]. Among the most significant advantages of virtualization implementation belong server consodilation bringing more efective usage of hardware, decrease in power consumption [2], clearer server management, etc. Other positives are effective access to backups, centralized management, and the possibility of testing the technologies being implemented. Utilization of virtualization is closely connected with architecture x86, or x86-64, and processor operation modes, in which the means of mapping the virtualization layer into its respective ring mode are significantly different. In architecture x86, protected mode utilizes 4 levels of permission: ring 0 (the highest level of permission) to ring 3 (the lowest level of permission). System memory is divided into segments, each of which is attributed to its respective ring. The processor then utilizes the given level of permission to determine what can and cannot be performed with the segment contents. The following article addresses problems concerning virtualization on architecture x86 and x86-64, and testing the efficiency of selected virtualization tools using type 1 hypervisor. In the conclusion individual tests are evaluated, and recommendations for possible optimal utilization of individual virtualization tools are summarized based on their usage.
II. VIRTUALIZATION ON X86
The problem with virtualizating architecture x86 is that platform x86 operating systems have been designed to run directly on computer's hardware [3] and they assume they can fully access said hardware using privileged instructions. Virtualization, however, requires that the hypervisor has access to the computer's hardware. Another problem is the existence of 17 special instructions that can create a conflict or a system error when not run in ring 0 [4]. Virtualization bypasses these specific features of x86 platform operating systems using its own internal algorithms. It is possible to extend the x86 architecture by 64bit addressing by adding a new mode named long mode. Abovementioned modes belong to a group labelled as legacy mode. Long mode consists of two submodes - a 64bit mode, which, among other things, supports the ability to address the whole 64bit virtual address space, and compatibility mode ensuring support of running 16
A. Type 1 hypervisor
Hypervisor, labeled as VMM (Virtul Machine Monitor), is in [1] defined as a software layer stored under virtual machines and physical hardware of the host computer that can create an abstraction for every VM, and which monitors the access of virtual machines to HW sources of the host machine. Type 1 hypervisor runs directly on the physical hardware without host operating system. It uses the highest level of permission, ring 0, to run. Hosted operating system then runs in a lower level of permission. It can directly communicate with hardware tools and is generally presumed to be safer than others, for in case of crash a virtual machine damages only its own closed working environment leaving other VM unaffected and capable of runing.
III. THE PRINCIPLES OF TESTING
To analyze individual virtualization tools and consequent recommendations for virtualization tools utilization, it is necessary to test the individual virtualization tools first. The tests were targeted on virtual machine’s key components, namely CPU, RAM, HDD and NIC (Network Interface Card). To ensure maximal objectivity during test evaluation, both host Microsoft Windows platform operating systems and unix platform operating systems were tested. For this reason, only tests and tools applicable to run on both platforms were selected to ensure maximal objectivity of testing and to maintain consistent conditions to the maximal possible degree. Each test was run one hundred times within the scope of testing, after each set of test a restart of the host system followed, and a sixty-second break to ensure the operating system stability after restart was kept. Host machine consisted of the same hardware configuration during all virtualization tools tests. Microsoft Windows Server 2012 R2 Standard Build 9600 and Debian GNU/Linux 7.4 Wheezy were chosen as platforms for the tests. All tests were performed on a computer with the following hardware configuration:
• Motherboard: ASUS P8H77-I, socket 1155, • processor: Intel Core i5-3550, 3.30 GHz, 6 MB L3
cache, (support IVT-x),
• hard disk: Seagate Desktop HDD 1TB, 7200 turn/min, 64 MB cache,
• system memory: Kingston HyperX black 16 GB, 1600 MHz, CL10,
• graphic card: Intel HD Graphics 2500 (650/1150 MHz),
• network card: Realtek 8111F, Gigabit LAN. All tests were first performed on both operating systems running directly on the host machine's physical hardware,
and then the tests were performed on each virtual machine according to the above described principles.
1) CPU test - π
The first test to measure processor performance was calculation of pi (π) to 8 388 608 (8M) decimals. For the calculation and data record System Stability Tester version 1.5.1 [5] testing software was used. System Stability Tester is free and open source software distributed under the terms of the GNU Public License version 2. Pi is calculated by using two system threads and Gauss-Legendre's algorithm [6]. To evaluate the results, shorter time means higher processor performance.
2) CPU test – Geekbench
The second test to determine processor performance was Geekbench software version 3.1.5 [7]. It is a new generation multiplatform processor benchmark with its own scoring system for performance in operation mode of one (Single-Core) and more cores separately, where higher score means higher processor performance. Geekbench offers a set of tests which simulate tasks a processor has to manage during a real operation system run on PC. Performed test tasks, from which evaluations are conducted, can be divided into three categories:
• Integer workloads: AES, SHA1, compression PNG and JPEG.
• Floating point workloads: Mandelbrot, Blur image, etc,
• Memory workloads: STREAM copy, STREAM add, etc.
For testing purposes, a version with the option of testing in 64bit mode was used.
3)HDD test – IOzone
Hard drive performance test was done using IOzone version 3-424 [9] tool. This tool was used to measure basic parameters of the hard drive. Measured parameters were the speed of a new file writing (including metadata) and the speed of reading an existing file within one file system. Tested file size was set to 4GB to eliminate file caching into operation RAM memory, as the measured results could be significantly biased. Results, listed in kB/s, were recorded directly from the IOzone tool output. To evaluate the results, the higher the transfer speed, the higher the hard drive performance.
4)NIC test – throughput
To measure network interface throughput, Iperf version 2.0.5-2 [9] was used. This tool works on a client-server principle. Iperf tool in server mode was run on a distinct TCP port on a server, which was the tested virtual machine. Repeated requests to measure throughput were then sent from a client - a factory set up Dell Optiplex 33019 connected to the host server via STP CAT5e cable, which is capable of transferring data at the speed of 1Gbits. The higher the throughput, the higher the network interface performance. Within the scope of the tool, the size of transferred data in one test iteration was set to 1000 MB.
5)NIC test – ping
Ping was measured using fping version 3.2-1 [11] program, which sends requests to a target station and measures the speed with which the station can respond. Just like more famous program ping. The fping program
was chosen for its ability to show results in microseconds, unlike the classic ping program. Default parameters the fping command uses for testing were used. The requests were send from the same external source as in the previous case and the results were recorded directly from the fping program output, to which applies that the higher the ping, the lower the network interface performance. B. Tested virtualization tools configuration
Each virtual machine had two processor cores, 4 GB RAM memory and 20GB HDD. File system ext4 was used for Debian GNU/Linux and file system NTFS was used for Windows Server. Other hardware configuration was different within the individual virtualization tools and it is directly specified at each case. All tested virtualization solutions were installed in its entirety, including additional drivers and software packages. If it was possible to use paravirtualization drivers, these were installed. Namely:
• Hyper-V (version 2012 R2): Integration components for Debian GNU/Linux using core 3.2. were installed in version 3.5 (from core version 3.3 and higher, these components are a part of the core itself [12]). Windows Server already contains intergation components. Windows Server was installed as a second generation virtual machine. Debian is not supported for second generation and was therefore installed as a first generation VM.
• VMware ESXi (version 5.5): Paravirtualization drivers for disc (PVSCSI) and for network interface (VMXNET3) were installed. Virtual machines were installed with VMware hardware version 10.
• Xen (version 4.4): Xen was compiled and installed from source codes and Debian Wheezy was used as dom0 because of required support of PVH. It was not necessary to add additional driver into Debian for linux virtualization (PV and PVH), they are already present directly in the core. However, it was necessary to compile the core of the host system with specific setting for enabling PVH for PVH virtualization, which is implicitly supported from core version 3.14 [14]. Paravirtualization driver (GPLPV) for hard drive and network interface in Windows systems were also installed. • KVM (version 1.1.2): Instalation was performed
from standard Debian repositories. In case of Debian virtualization, paravirtualization drivers VirtIO were used for hard drive and netwrok interface. Digitally signed paravirtualization drivers developed by Red Hat were used for virtualization of Windows Server.
IV. THE TEST RESULTS
The tests were first performed in MS Windows Server 2012 R2 Standard and Debian GNU/Linux 7.4 Wheezy operating systems, which ran directly on the tested host machine hardware. The number of processor cores the systems could use was limited to two and size of RAM memory was limited to 4 GB to enable comparison in
performance of virtualized and classically operated operating system. Only one virtual machine was running at a time. The following abbreviations are then used in tables with results:
• SD - the standard deviation • Q1 - the first quartile • Q3 - the third quartile • PV - paravirtualization
• HVM - full virtualization hardware support • PVHVM - HVM to support paravirtualization • PVH - paravirtualization inside HVM container A. CPU tests
1)The calculation of π
The first test, which was focused on processor performance, was the calculation of pi (π) to 8 388 608 (8M) decimals. As it was mentioned above, the software used for the calculation was System Stability Tester, in which π is calculated by using two system threads and Gauss-Legendre's algorithm. If we use the native run of operating system as a base value, it is apparent from the results that time increase in calculation by virtualized machines does not exceed 32% for Debian virtualization and 13% for Windows Server virtualization. In this test the tool that showed the best results (lowest calculation time) was KVM for Debian virtualization and ESXi for Windows Server virtualization.
In terms of the host machine utilization all virtualization solutions were close to 50 % (49-51%). Only Xen PVH virtualization loaded the physical machine in an interval from 45 to 50 %.
TABLE I. THE CALCULATION OF Π RESULTS Tools Average Max. Min
. SD Median Q1 Q3 Debian OS virtualization Debian 26.87 27.0 27.1 0.1 26.9 26.8 26.9 Hyper-V 29.06 29.3 29.6 0.1 29.1 29.0 29.1 ESXi 28.51 28.6 28.9 0.0 28.5 28.5 28.5 KVM 27.88 28.7 28.5 0.2 27.9 27.8 27.9 Xen-PV 34.79 34.9 35.1 0.1 34.8 34.7 34.8 Xen-PVH 35.50 35.9 33.5 0.3 35.5 35.3 35.8 Windows OS virtualization Windows 29.74 30.2 30.2 0.1 29.7 29.7 29.7 Hyper-V 30.96 31.7 31.6 0.2 30.9 30.9 31.0 ESXi 30.48 31.0 30.7 0.2 30.5 30.4 30.5 KVM 32.05 32.5 32.8 0.3 32.0 31.8 32.2 Xen PVHVM 33.69 34.4 34.6 0.2 33.7 33.6 33.8 2) CPU test - Geekbench Single-Core
Based on the below shown results it can be concluded that all tested tools show very similar results during CPU testing via Geekbench tool in Single-Core mode.
TABLE II. GEEKBENCH SINGLE-CORE RESULTS
Tools Average Max. Min. SD Median Q1 Q3 Debian OS virtualization Debian 3268.90 3276 3262 3.176 3269.0 3267.0 3270.5 Hyper-V 3238.60 3268 3165 20.3 3241.5 3235.8 3250.3 ESXi 3222.25 3264 3172 27.13 3239.0 3197.5 3244.3 KVM 3008.20 3017 3001 3.881 3008.0 3005.8 3010.3 Xen-PV 2958.20 2975 2947 5.963 2958.0 2954.8 2960.3 Xen-PVH 3164.65 3169 3161 2.104 3165.0 3163.0 3165.3 Windows OS virtualization Windows 3137.45 3145 3131 3.528 3138.0 3135.5 3139.3 Hyper-V 2969.50 3042 2888 38.64 2976.0 2963.0 2988.5 ESXi 3053.00 3075 3030 12.17 3055.0 3043.5 3063.0 KVM 3115.60 3141 3094 11.98 3116.0 3109.8 3121.0 Xen PVHVM 3097.25 3102 3092 2.981 3096.5 3095.0 3100.0 If we use native run of virtualized operating system as etalon, we will see that the maximal decrease in performance of the virtualized OS in contrast with the native one can be seen in Debian virtualization Xen in mode PV , which shows decrease in performance in contrast with the native OS by 10% and in Windows server virtualization it can be seen in Hyper-V with the decrease by almost 5%. Interesting results for Debian are shown by Hyper-V, and for Windows Server by KMV. The decrease in performance in this test was whole 1% for both. Evaluations Q1 and Q2 and the median value, which are close to the average value, then show the stability of individual virtualization tools.
All virtualization tools caused load of the host machine of about 15-25% during this test. This large variance is caused by Geekbench test profile, which consists of many small tasks.
3) CPU test - Geekbench Multi-Core
Based on the below shown results, it can be concluded that all tested tools show very similar results during CPU testing via Geekbench tool in Multi-Core mode as in the previous measuring. Maximal decrease in performance of the virtualized OS in contrast with a native one was noted in Xen in PV mode in Debian, which shows decrease in performance in contrast with native OS by 11%, and in Windows Server it is Hyper-V with the decrease by almost 7%. Hyper-V for Debian virtualization and KMV for Windows Server show the best results, in which the decrease in performance in this test was whole 2% in both. In Multi-Core mode a slight deterioration in results is apparent and it can be said that virtualization in Single-Core mode is by approximately 3% more effective than in Multi-Core mode. Host machine CPU load during this test was almost identical with the vlaues of 30-35%.
TABLE III. GEEKBENCH MULTI-CORE RESULTS
Tools Average Max. Min. SD Median Q1 Q3 Debian OS virtualization Debian 5885.9 5899 5871 6.6 5888 5882 5891 Hyper-V 5727.5 5774 5627 41.1 5745 5721 5753 ESXi 5708.8 5771 5549 70.4 5750 5653 5758 KVM 5371.6 5395 5337 15.8 5372 5361 5383 Xen-PV 5291.6 5327 5268 19.2 5289 5274 5303 Xen-PVH 5664.5 5745 5617 46.0 5641 5620 5700 Windows OS virtualization Window s 5639.4 5652 5607 10.5 5641 5632 5646 Hyper-V 5254.4 5379 5109 68.7 5291 5195 5300 ESXi 5548.7 5573 5510 16.4 5552 5537 5560 KVM 5586.2 5596 5569 6.8 5585 5581 5592 Xen PVHVM 5572.3 5596 5535 14.2 5575 5570 5580 B. HDD test
1)HDD test - IOzone – write
The highest speed of writing in the result of speed of writing test during virtualization via IOzone program was noted in virtualization XEN PV, which shows higher values than the native Debian system. This can be explained by the support of special instructions on the level of XEN virtualization tool. For Windows virtualization, the best values were noted in virtualization via KVM, which in contrast with native Windows server shows decrease by less than 1%.
TABLE IV. TESTS HDD WRITE RESULTS Tools Averag
e Max. Min. SD Median Q1 Q3 Debian OS virtualization Debian 208889 211563 206061 1593 208693 207567 209834 Hyper-V 202889 207935 195736 3258 203853 200890 205234 ESXi 180843 187915 164597 5702 181293 180025 184192 KVM 204964 211714 194334 3950 204616 202898 208426 Xen-PV 229185 232940 219347 3153 230001 228758 230958 Xen-PVH 228420 234059 224464 2646 228557 226293 230219 Windows OS virtualization Windows 198442 211945 188573 5466 198100 195536 200827 Hyper-V 174122 184210 161244 6303 175985 169360 179522 ESXi 178387 184539 171579 3441 179216 175860 180446 KVM 196164 208148 174681 7536 198153 194924 200359 Xen PVHVM 112882 130453 103803 8217 110375 105537 118555 Despite Windows Server virtualization using Xen PVHVM requiring special paravirtualization drivers for hard drive, which were installed during testing, its
performance is interestingly low. The decrease in contrast with native Windows server is by more than 43%.
1) HDD test - IOzone – reading
Solutions Xen PV and Xen PVH confirmed excellent results within the reading from hard drive test. The best results were noted in solution Xen PV in Debian virtualization, which values were almost identical to the values of a native Debian OS. The worst results were noted in virtualization using the solution ESXi. Here, the decrease in contrast with native Debian was about 20%.
TABLE V. TESTS HDD READING RESULTS
Tools Average Max. Min. SD Median Q1 Q3 Debian OS virtualization Debian 201255.0 203079 198591 1152 201444 200644 202169 Hyper-V 195300.0 200054 179726 5231 196904 194686 198720 ESXi 162805.0 169854 155939 3836 163182 159726 164926 KVM 199964.0 203426 199822 3950 197898 189334 199616 Xen-PV 201140.0 203664 195954 1930 201363 199945 202707 Xen-PVH 201260.0 203727 196790 2045 201743 201047 202358 Windows OS virtualization Windows 197750.0 205239 191297 3694 198438 195166 200728 Hyper-V 176790.0 185156 154922 6568 177364 175583 180125 ESXi 162805.0 169854 155939 3836 163182 159726 164926 KVM 192121.0 202684 177363 8106 195236 185119 198362 Xen PVHVM 182876.0 198763 159758 10493 184904 173735 191033 In Windows Server virtualization the results are significantly balanced with the exception of solution ESXi, which shows decrease by 18%. The best results were noted in the solution KVM for Windows Server virtualizaion, which fell behind the native Windows Server by almost 2%. During both the reading and the writing test the load of host machine ranged between 5-12%. The best did Hyper-V, which loaded the physical machine during the test by about 3%.
C. NIC test
1) Network throughput – Iperf
In the network troughput test we get some very interesting results for network throughput efficiency for OS Debian virtualized tools. All tools, with the exception of ESXi, reached 98% network interface pefromance of natively running system. Statistics for Windows Server virtualization are slightly worse. Only two tools, XEN PVHVM and Hyper-V, reached 90% performance of the native system. The most interesting result is seen in a very bad performance of ESXi, namely in Windows Server virtualization, where the measured values show only 54% of native system performance and when using it on Debian virtualization, it reached almost 80%. The biggest load of host machine was seen while running KVM, which required 37% of the host machine CPU performance in Windows Server and 25% of the host machine CPU performance in Debian. All three tested Xen virtualizations caused load lower than 10%. Other tools required about 15% of CPU performance.
TABLE VI. TESTS NICNETWORK THROUGHPUT RESULTS Tools Average Max. Min. SD Median Q1 Q3
Debian OS virtualization Debian 942.00 942.0 942.0 0.0 942.0 942.0 942.0 Hyper-V 927.10 941.0 783.0 36.5 940.0 939.0 940.3 ESXi 750.55 771.0 703.0 17.5 751.0 743.0 767.0 KVM 935.10 942.0 891.0 12.7 939.0 937.8 941.0 Xen-PV 939.60 941.0 937.0 1.4 940.0 938.8 941.0 Xen-PVH 940.15 942.0 933.0 2.1 941.0 940.0 941.0 Windows OS virtualization Windows 941.00 941.0 941.0 0.0 941.0 941.0 941.0 Hyper-V 856.90 871.0 836.0 9.7 857.5 852.3 863.3 ESXi 510.45 571.0 421.0 37.8 514.5 490.3 538.0 KVM 736.80 750.0 719.0 8.4 738.0 732.0 743.3 Xen PVHVM 940.15 942.0 933.0 2.1 941.0 940.0 941.0 1) Network response – fping
In the network response test we get some very interesting results for OS Debian virtualized tools. All tools, with the exception of ESXi, reached 98% of the network interface pefromance in natively running system. ESXi tool results reached 80% of the native OS Debian run.
Windows Server virtualization results show performance comparable with native operating system of about 90%. Virtualization using ESXi is the exception with the worst times of response closely above 50% just like in Debian virtualization.
TABLE VII. TESTS NICNETWORK RESPONSE RESULTS Tools Average Max. Min. SD Median Q1 Q3
Debian OS virtualization Debian 0.133 0.15 0.12 0.01 0.13 0.13 0.14 Hyper-V 0.287 0.44 0.22 0.05 0.27 0.26 0.32 ESXi 0.962 1.16 0.75 0.09 0.95 0.91 1.02 KVM 0.332 0.35 0.28 0.02 0.34 0.33 0.34 Xen-PV 0.283 0.48 0.15 0.07 0.28 0.26 0.28 Xen-PVH 0.279 0.32 0.27 0.01 0.28 0.27 0.28 Windows OS virtualization Windows 0.196 0.23 0.17 0.02 0.19 0.19 0.20 Hyper-V 0.369 0.41 0.32 0.03 0.37 0.35 0.40 ESXi 1.064 1.33 0.92 0.09 1.04 1.03 1.11 KVM 0.46 0.48 0.44 0.01 0.46 0.45 0.47 Xen PVHVM 0.314 0.40 0.21 0.04 0.33 0.29 0.33
V. EVALUATION OF THE TEST RESULTS A. CPU test
In total three test were performed to investigate the virtualized systems CPU performance. For linux virtualization it is positively recommended to use Microsoft Hyper-V virtualization, which consistently showed excellent results in all conducted tests and its decrease in performance was noted to be up to 7% in contrast with the native operating system Debian run. In case of a bigger isolation of individual virtual machines request, it is possible to use ESXi by VMware. This tool did not show the best values during each test focused on CPU performance, but its results did not show any extremes, and efficiency of this virtualization was around 96% of the native operating system Debian.
For virtualization of the MS Windows family systems it is not easy to recommend a virtualization solution. Both ESXi and KVM tools did very well, their results reached about 97% of the native operating system Windows. Hyper-V had shown good performance in hard information power while calculating π, but its results in complex Geekbench tests were sub-standard.
B. HDD test
To test the speed of writing on and reading from hard drive in virtual machines, two tests were performed. Virtualization tools, which did well in these tests can be recommended for, for example, virtualizing database servers, in which the speed of I/O operations is critical. For Debian virtualization, XEN-PV and XEN-PVH virtualization did unexpectedly well. KVM tool lost only a few percent in contrast with the native speed and can also be recommended. Concerning the speed of data writing and reading, the best candidate for MS Windows virtualizations seems to be KMV, which beat both Hyper-V and PHyper-VHHyper-VM virtualizations by Xen rather significantly. ESXi, which passed the speed of writing test, failed the speed of reading test, and cannot therefore be recommended as an appropriate solution.
C. NIC test
Network interface performance was examined in two tests. Defining NIC speed parameters were throughput and ping speed on request. Network interface performance is critical for, for example, modern cloud solutions, which have high demands for network connection quality. In Debian virtualization test several solutions did almost as same as the speed of network interface on the native operating system and all tested solutions can therefore be recommended.
Only one tested Windows system virtualization tool managed to match the speed of the native network interface and that was Xen PVHVM. Second was Hyper-V solution, which lost nearly 10% of network card speed. Definitely the worst solution surprisingly proved to be ESXi and can not be in any way recommended for virtualizing systems, which have high requirements for network interface performance.
VI. CONCLUSION
tests of chosen virtualization tools utilizing type 1 virtualization with focus on the speed of key components, such as CPU, RAM, HDD and NIC, and to map the
development in virtualization technologies. Acquired data also serve as a clue for choosing the optimal virtualization solution for specific utilization. Results, that can be deemed valuable, are the ones showing comparable performance of virtualized machine with natively running operating system. Efficiency of CPU usage, work with hard drive, and optimalization of work with NIC did not increase since the last measuring [13], [14]. Old stand-bys in this field are platforms KVM and ESXi, which show standard values without any major long-term deviation to both positive an negative numbers. A surprise was virtualization tool Hyper-V by Microsoft, which showed above average values in tests, and in many parameters, mainly with virtualization of operating system Debian, it was a leader in all tested tools.
For virtualization of systems with a Linux core tools KVM and Hyper-V can generally be recommended, as they showed minimal decrease of 7% in performance in all test in contrast with the native operating system. For virtualization of operating systems on Microsoft platform, tools KVM or ESXi can be recommended.
Generally, the test results can be concluded by a statement that none of the tested tools did not extremely fall behind the others and in their utilization it is necessary to take into account whether the virtualized machine will provide specific services oriented on, for example, using NIC or HDD or whether the virtualized machine will provide a wide variety of services. Finally, the choice of the virtualization tool is also affected by the user interface for instalation and administration of the utilized system, which cannot be evaluated objectively, but it plays a significant role.
ACKNOWLEDGMENT
This work and contribution is supported by the project IN2 CZ.1.07/2.2.00/28.0272 financed from European Social Funds.
REFERENCES
[1] G. Eason, B. Noble, and I. N. Sneddon, “On certain integrals of Portnoy, M.. Virtualization Essentials. Indianapolis: Sybex, 2012. ISBN 978-1118176719.
[2] Horalek, J., Suba, P., Sobeslav, V., Matyska, J.: Energy Efficiency Measurements of Data Center Systems. (2014) 10th International Conference - ELEKTRO 2014, pp. 41-45.41-45.
[3] Hess, K., ,Newman, A.: Practical Virtualization Solutions: Virtualization from the Trenches. Boston: Prentice Hall, 2009. ISBN 978-0137142972.
[4] Dittner, R., Rule, D.: The Best Damn Server Virtualization Book Period: Including Vmware, Xen, and Microsoft Virtual Server. Burlington: Syngress, 2007. ISBN 978-1597492171.
[5] System Stability Tester. [online]. 2007, 26.6.2012 [cited by 2014-07-03], available from: http://systester.sourceforge.net/
[6] P. Eymard, P. Lafon. The number [pi]. Providence, R.I.: American Mathematical Society, 2004, 322 p. ISBN 08-218-3246-8. [7] Primate labs inc. Geekbench 3 [online]. 2013 [cited by
2014-07-03], available from: http://www.primatelabs.com/geekbench/ [8] Freecode. IOzone [online]. 2008 [cited by 2014-07-03], available
from: http://freecode.com/projects/iozone/
[9] Iperf [online]. 2008 [cited by 2014-07-03], available from: http://iperf.sourceforge.net/
[10] T. Dzubin. Fping [online]. 2007 [cited by. 2014-07-03], available from: http://fping.sourceforge.net/
[11] A. Finn, P. Lownds, M. Luescher, D. Flynn. Windows Server 2012 Hyper-V Installation and Configuration Guide. Indianapolis: Sybex, 2013. ISBN 978-1118486498.
[12] M. Larabel. Xen PVH Is Landing For The Linux 3.14 Kernel. Linuxtoday.com [online]. 2014 [cited by 2014-04-18]. available from: http://www.linuxtoday.com/developer/Xen-pvh-is-landing-for-the-linux-3.14-kernel.html
[13] Horalek, J., Matyska, J., Sobeslav, V.:Performance comparison of selected virtualization platforms. (2013) CINTI 2013 - 14th IEEE International Symposium on Computational Intelligence and Informatics, Proceedings, art. no. 6705215, pp. 327-332.327-332. [14] Suba, P., Horalek, J., Hatas, M.: Comparison of technologies for
software virtualization. (2011) Recent Advances in Applied and Biomedical Informatics and Computational Engineering in Systems Applications - AIC'11, BEBI'11, pp. 160-165.160-165.