More and more organizations are using virtualization techniques to shed flab
in their data center and also cut on power costs. And there is no dearth of
vendors selling virtualization solutions. The market has become so competitive
that you can literally pick and choose the solution that best fits your needs.
In our previous articles, we have talked about how virtualization can help you
save costs, space and power. Now, as the number of virtualization deployments
have increased, so has the complexity involved in managing them. It's imperative
that you know how different virtualization platforms perform and what you should
look for while choosing them.
Here, we test the top three different virtualization platforms: VMware
Infrastructure 3.5, MS Hyper-V and Citrix Xen Server 5 to help you make an
informed decision.
Test setup
For testing we used the latest Intel Dunnington Server having 4 processors with
6 cores each and 16 GB RAM. For benchmarking we used CINEBENCH 10 64-bit, POV
Ray 64-bit and Linpack. We ran these benchmarks on Windows 2003 Enterprise
64-bit OS which was installed virtually on different platforms. We tried to form
a common ground for comparing these platforms based on the number of processor
cores in each platform.
Direct Hit! |
Applies To: Data center |
We found out that MS Hyper-V and VMware Server allow only upto 4 cores per
processor whereas Xen Server allows allocation of 8 cores per virtual machine.
So, we took 4 cores as the standard across all platforms. But we also tested for
8 cores with the Xen Server, so that we get to know the difference in
performance.
For managing VMware server and Xen Server, we used a Core2 duo machine with 1
GB RAM where we installed clients for both, Xen Center for Xen Server and VMware
infrastructure management console for VMware Server. The management machine and
the Server were connected over Gbps network. We also tried running SunGard on
the virtual machine of different platforms, but it didn't run.
Test results
CINEBENCH 10 64-bit: Initially we started the test with CINEBENCH 10 which
measures the performance of processor and graphic cards.
We first installed Windows Server Datacenter 2008 along with Hyper-V on the
Dunnington server. Then we created a new virtual machine with 4 cores, 15 GB RAM
and 20 GB HDD. On the virtual machine we installed Windows Server 2003
Enterprise 64-bit Edition.
Once the OS installation finished we installed CINEBENCH 10 on the virtual OS
and then ran the benchmark. With one core on Hyper-V it gave 3052 CB-CPU and for
4 Cores it gave 10565 CB-CPU. Next we installed Xen Server on Dunnington server
and with its client tool, we created a new virtual machine. We allocated the
same resources as for the previous virtual machine, and installed Windows Server
2003 Enterprise 64-bit. Again we installed CINEBENCH 10 and ran the benchmark.
For one core it gave a score of 3028 CB-CPU which is slightly less than Hyper-V
but on 4 cores it gave 11057 which is higher than Hyper-V. We repeated the same
process with VMware Server, and with one core, CINEBENCH 10 gave 3079 CP-CPU
which is a bit higher than both Hyper-V and Xen Server. With 4 cores it gave
11103 CB-CPU with is higher than both Hyper-V and Xen Server. We also tested
CINEBENCH 10 on 8 cores with Xen Server and found that the score were really
amazing. It gave a score of 19458 CB-CPU which is far greater than what 4 cores
can provide.
POV Ray: Next we used the ray tracing program POV Ray which is again
used for CPU benchmarking. It uses the ray tracing rendering technique to
calculate how may image pixels are rendered per second by simulating how light
travels in real world. For running this benchmark we followed the same procedure
as that for running CINEBENCH 10.
We installed the host OS and then created a guest with Windows Server 2003
Enterprise OS, and allocated the same resources as that while running CINEBENCH
10. In Hyper-V it rendered 118.18 PPS over 147456 pixel which took a total of
1247.75 seconds.
In the case of VMware Server it rendered 116.42 PPS over the same amount of
pixels which is less compared to Hyper-V. And also the total time taken by
VMware Server is 1266.55 seconds which is more than Hyper-V. On Xen Server, POV
Ray rendered 119.68 PPS over 147456 pixels which is more as compared to the two
other platforms and took 1233.36 seconds which is also lesser than both. But on
8 cores for Xen server, it rendered only 119.60 PPS over the same amount of
pixels.
The reason behind this is that POV Ray used a maximum of 3 cores, and that
too not simultaneously.
Linpack: Next we ran Linkpack which is the toughest test for any
server. It measures a system's floating point computing power by making the
system solve an N by N linear equation (ie Ax = b). It calculates the maximum
number of GFlops that can be generated.
The setup for running this benchmark is similar to the previous benchmarks.
On Hyper-V it gave 28.6 GFlops and on VMware server it gave 29.93 GFLops which
is good as compared to Hyper-V. But for achieving these results we had to do
some tweaking to its input file. On the Xen server Linpack gave 33.84 GFlops
which is better than the other two.
Now with 8 cores the figure was really amazing. It gave 64.25 GFlops which is
comparable to any physical server score.
We reviewed the Harpertown server (January 2008 issue) with two 3 Ghz
quad-core processors, 16 GB RAM and two 200 GB SATA hard disks connected through
a PCI-X RAID controller. On Harpertown, Linpack gave 65 GFlops which is a bit
higher than this virtual machine created on Xen Server.