www.phoronix.com/scan.php?page=article&item=ubuntu-1510-virt&num=1
Our latest benchmarks of Ubuntu 15.10 are looking at the performance of this latest Linux distribution release when comparing the performance of guests using KVM, Xen, and VirtualBox virtualization from the same system.
The tests were all done from an Intel Xeon E5-2687W v3 + MSI X99S SLI PLUS system with 16GB of DDR4 system memory, 80GB Intel M.2 SSD, and AMD FirePro V7900 graphics. Once running our disk and processor focused benchmarks on this Ubuntu 15.10 host system, the "bare metal" results were then compared to a KVM guest setup via virt-manager using the Ubuntu Wily packages, then using the Xen 4.5 packages present on Ubuntu 15.10 with again testing the same Ubuntu 15.10 guest with virt-manager, and then lastly testing the Ubuntu 15.10 guest under VirtualBox 5.0.4 as available via the package archive.
The host system and all guests were using Ubuntu 15.10 64-bit with the Linux 4.2.0-16-generic kernel, Unity desktop, X.Org Server 1.17.2, an EXT4 file-system, and GCC 5.2.1 as the code compiler. All system settings remained the same during the testing process.
When creating the KVM/Xen/VirtualBox guests, 38GB of virtual storage on the Intel SSD, 8GB of RAM and all 20 CPU threads (ten core CPU + HT) were made available to the virtual machine under test. As each platform has near endless tunable possibilities, the testing was done with the default settings. If there's enough interest and premium support from Phoronix readers, I can look at putting out some "tuned" VM comparisons.
All of the benchmarks for this article were carried out via the open-source Phoronix Test Suite benchmarking software.
+++++++++++++++++++++++++++++++++++++++++++
http://www.phoronix.com/scan.php?page=article&item=ubuntu-1510-virt&num=2
First up are some of the disk benchmarks for the host system and then the KVM/Xen/VirtualBox guests
With VirtualBox's reported result outperforming the bare metal system, it looks like VirtualBox in its default configuration for this SQLite database benchmark isn't fully honoring fsyncs to the disk compared to KVM and Xen with the performance being too good. While KVM was the slowest, if you're doing a lot of write-intensive work in your VMs, you are better off letting the VM access a raw partition rather than just setting up a virtual disk.
With the random and sequential writes, VirtualBox was reported to be faster than KVM and Xen but again may be a matter of its different behavior.
When it comes to sequential reads, VirtualBox was slower than the competition.
Dbench is another disk benchmark showing VirtualBox's different behavior of apparently not writing everything out to the disk during testing with the reported performance being too good.
+++++++++++++++++++++++++++++++++++++++++++
http://www.phoronix.com/scan.php?page=article&item=ubuntu-1510-virt&num=3
KVM was edging past Xen and VirtualBox for the CompileBench test.
With the Parboil OpenMP CPU test, the performance of KVM and Xen effectively matched that of the host system with minimal overhead in having access to all 20 CPU threads and having no other tasks active on the host system during testing. While VirtualBox was configured with the same hardware access, it was much slower than KVM and Xen.
The Rodinia OpenMP test was also showing VirtualBox at a similar disadvantage.
+++++++++++++++++++++++++++++++++++++++++++
http://www.phoronix.com/scan.php?page=article&item=ubuntu-1510-virt&num=4
KVM was faster than Xen at the MAFFT scientific test while VirtualBox 5.0 was nearly twice as slow.
On this Haswell Xeon system, KVM was faster than Xen and VirtualBox.
Xen scored a win past KVM when it came to the GraphicsMagick image resizing test.
There wasn't a huge performance difference in the largely single-threaded Himeno pressure solver.
+++++++++++++++++++++++++++++++++++++++++++
http://www.phoronix.com/scan.php?page=article&item=ubuntu-1510-virt&num=5
KVM and Xen did fine on the timed kernel compilation benchmark while VirtualBox was much slower.
It seems nearly across the board that VirtualBox 5.0 performs poorly with the multi-threaded tests even though it too had access to all of the Intel Xeon v3's threads.
There's only small performance differences in the multimedia encoding tests.
Overall, Xen and KVM were performing well compared to the host's performance. Xen tended to have the most wins while KVM generally came in right behind and the two were very competitive with one another, either are great solutions for Linux virtualization depending upon your specific needs. The VirtualBox results, however, were a bit concerning at least for its stock configuration with it performing poorly near universally in the multi-threaded benchmarks and the behavior in the disk benchmark also being potentially problematic.
If you would like to support our continued Linux benchmarking work at Phoronix, please consider subscribing to Phoronix Premium.