August 22, 2009 at 11:30A small test of the need (or unnecessary) VT when working with Virtualbox
The motherboard died from the main home file washer, I had to buy a new one. At the same time, the processor changed (there was a PDC E2160, became Celeron DC E3200).
In addition to home file storage and a router, a couple of web servers in Virtualbox are spinning on the server (guests are Linuxs). And new Celerons, unlike previous cheap Intel processors, have a function of hardware virtualization (VT). I wanted to check if it was useful for using virtualbox. Because Google gave links only to very old tests, in which the inclusion of VT slowed down.
The main difficulty was to find some small benchmark for Linux. I know that a common way to measure performance is to build the kernel, but I wanted something simpler.
In search of "simpler" went to procbench . Just a very simple program with several tests. It is clear that synthetics has a weak relation to real life, but in this case I just had to compare the performance in parrots.
To measure the performance of the disk subsystem, I used bonnie ++ (available in the distribution), with the default settings (bonnie).
Host: Celeron Dual-Core E3200 @ 2.4GHz, 3.5GB RAM, Windows Web Server 2008 x64, VirtualBox 3.0.4
Guest: Debian Lenny, 1GB RAM, vbox guest additions installed.
The first series of tests - VT is disabled, Nested Paging is disabled.
The second series of tests - VT is on, Nested Paging is on.
100 procbench starts were made, then the average result was considered.
bonnie was launched five times, the results are also average. Between starts, the host rebooted - to completely clear the cache (As it turned out, the host caches information for the guest. For example, if I copied the file to / dev / null, then rebooted the guest, after the reboot the file was copied in a second - i.e., disk cache not reset by guest reboot).
www.textuality.com/bonnie/advice.html - a little legend
The average performance increase in processor tests was 6.51%. Nice, but not critical.
But the speed of disk operations has grown very noticeably - there is either an increase in the speed of disk operations, or a decrease in processor load when performing disk operations. Or both.
In general, my doubts about the need for VT are dispelled. It is clear that it should have provided an increase in productivity, but now I can at least see what and how much I get.
In addition to home file storage and a router, a couple of web servers in Virtualbox are spinning on the server (guests are Linuxs). And new Celerons, unlike previous cheap Intel processors, have a function of hardware virtualization (VT). I wanted to check if it was useful for using virtualbox. Because Google gave links only to very old tests, in which the inclusion of VT slowed down.
The main difficulty was to find some small benchmark for Linux. I know that a common way to measure performance is to build the kernel, but I wanted something simpler.
In search of "simpler" went to procbench . Just a very simple program with several tests. It is clear that synthetics has a weak relation to real life, but in this case I just had to compare the performance in parrots.
To measure the performance of the disk subsystem, I used bonnie ++ (available in the distribution), with the default settings (bonnie).
Host: Celeron Dual-Core E3200 @ 2.4GHz, 3.5GB RAM, Windows Web Server 2008 x64, VirtualBox 3.0.4
Guest: Debian Lenny, 1GB RAM, vbox guest additions installed.
The first series of tests - VT is disabled, Nested Paging is disabled.
The second series of tests - VT is on, Nested Paging is on.
100 procbench starts were made, then the average result was considered.
bonnie was launched five times, the results are also average. Between starts, the host rebooted - to completely clear the cache (As it turned out, the host caches information for the guest. For example, if I copied the file to / dev / null, then rebooted the guest, after the reboot the file was copied in a second - i.e., disk cache not reset by guest reboot).
CPU
Virtualization is disabled:
| 1.ctest1 (multiplication 1x10 ^ 9 times in nested loop) | 1339 ms |
| 2.ctest2 (Ackermann's function of (3, 12)) | 3979 ms |
| 3.ctfibo (Fibonacci number of 42) | 5167 ms |
| 4. ctheapsort (Heap sort on 40mb random data) | 4223 ms |
| 5.ctmatrix (Matrix multiplication) | 3734 ms |
| 6. ctrandom (Random number generation) | 1975 ms |
Virtualization enabled:
| 1.ctest1 (multiplication 1x10 ^ 9 times in nested loop) | 1282 ms |
| 2.ctest2 (Ackermann's function of (3, 12)) | 3647 ms |
| 3.ctfibo (Fibonacci number of 42) | 4945 ms |
| 4. ctheapsort (Heap sort on 40mb random data) | 3673 ms |
| 5.ctmatrix (Matrix multiplication) | 3556 ms |
| 6. ctrandom (Random number generation) | 1889 ms |
Performance Comparison (Percentage Growth):
| 1.ctest1 (multiplication 1x10 ^ 9 times in nested loop) | 4.26% |
| 2.ctest2 (Ackermann's function of (3, 12)) | 8.34% |
| 3.ctfibo (Fibonacci number of 42) | 4.30% |
| 4. ctheapsort (Heap sort on 40mb random data) | 13.02% |
| 5.ctmatrix (Matrix multiplication) | 4.77% |
| 6. ctrandom (Random number generation) | 4.35% |
HDD
Virtualization is off:
Version 1.03d ------ Sequential Output ------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K / sec% CP K / sec% CP K / sec% CP K / sec% CP K / sec% CP / sec% CP
banshee.xxxxxxxx 2G 22737 86 34572 61 18864 41 30355 84 71007 76 562.9 21
------ Sequential Create ------ -------- Random Create --------
-Create-- --Read --- -Delete-- -Create-- --Read --- -Delete--
files / sec% CP / sec% CP / sec% CP / sec% CP / sec% CP / sec% CP
16 6218 97 +++++ +++ 8799 99 6651 99 +++++ +++ 7900 1
banshee.xxxxxxxxx.xx, 2G, 22737.86.34572.61.18864.41.31.30355.84.71007.76.562.9,21.16.6218.97, +++++, +++, 8799.99, 6651.99, +++++, +++, 7900.1Virtualization enabled:
Version 1.03d ------ Sequential Output ------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K / sec% CP K / sec% CP K / sec% CP K / sec% CP K / sec% CP / sec% CP
banshee.xxxxxxxx 2G 43997 86 33739 13 29359 10 41485 79 88934 15 406.5 6
------ Sequential Create ------ -------- Random Create --------
-Create-- --Read --- -Delete-- -Create-- --Read --- -Delete--
files / sec% CP / sec% CP / sec% CP / sec% CP / sec% CP / sec% CP
16 +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++
banshee.xxxxxxxxx.xx, 2G, 43997.86.33739,13,29359,10,41485,79,88934,15,406.5,6,16, +++++, +++, +++++, ++ +, +++++, +++, +++++, +++, +++++, +++, +++++, +++
www.textuality.com/bonnie/advice.html - a little legend
Conclusion
The average performance increase in processor tests was 6.51%. Nice, but not critical.
But the speed of disk operations has grown very noticeably - there is either an increase in the speed of disk operations, or a decrease in processor load when performing disk operations. Or both.
In general, my doubts about the need for VT are dispelled. It is clear that it should have provided an increase in productivity, but now I can at least see what and how much I get.