Virtualization Platforms Compared

We put to test three popular virtualization platforms: Xen Server, VMware and MS Hyper-V to help you pick the best one for your need

Monday, December 01, 2008

Print

Comment

Email

Digg

Del.icio.us

Reddit

Twitter

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.

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.

Page(s)   1