Result

The result was that the virtualized installation on vSphere 5.5 was roughly 12% slower than the natively installed environment. The authors also explain that roughly 5% of these 12% can be explained through the limitation of vSphere to 64 virtual CPUs (vCPU) which are mapped to hardware threads. As the hardware has 60 cores with 120 hardware threads (as Hyper-threading is enabled), the virtual environment can only use a little more than half the threads.

The authors also explained that when disabling the hyper-threading on a native installation, the difference is roughly these 5%. So the advantage of native installations could be smaller when either having more vCPUs available in vSphere or by disabling hyper-threading on the native installation (which is of course not really an useful idea).

So from my perspective these numbers sound promising. As soon as the limitations of 64 vCPUs and 1TB Memory per Virtual Machine are removed the virtualization might also be an option for production environments. For development and test environments these numbers are already really good news.

Sources

• VMware VROOM! Blog: First Certified SAP BW-EML Benchmark on Virtual HANA 
• VMware VROOM! Blog: Virtual SAP HANA Achieves Production Level Performance 
• SAP HANA on VMware on vSphere Best practices guide 
• SAP HANA virtualized - Overview