Virtualization, Hardware, Operating Systems, Tools, IT Service Management and more... Jens Wagener´s Blog

lies, damn lies and benchmarks....

Note how they used a 4 core system with a 1 cpu virtual machine and compared it to a non-smp redhat kernel (effectively limiting it to 1/4th the system). Sounds fair, right? Not exactly.

This gives an advantage to the virtual environment since 1 cpu can be used to host the VM while the other 3 cores can be used for the ESX overhead (i/o and such). With the physical machine, the OS only has 1 cpu core to work with.

I would be more interested to see a dual proc (single core) system with a 2 cpu VM vs a smp redhat kernel (thus they both get the same CPU resources to play with). Another option would be to show a single physical box (smp kernel) vs 4 VM's, each with 1 cpu.

That said, we use a lot of ESX in our environment. It has its place. We pile 10+ VM's on a 4 core box. In those cases, we dont need performance, just the ability to save money on hardware. In these cases, the $7k for the ESX software is not a big deal compared to the cost in hardware savings (rack space, power, servers, switch ports, etc, etc). If we need performance, we use things like solaris zones or just give the OS its own box.

Anyway, I find that this particular setup is a big stacked in VMWARE's favor. Given the results of the test, I'm not as impressed as they think I should be.

Gesendet von John am Februar 23, 2008 at 05:39 PM CET #