Performance should be generally the same, regardless of 32 or 64 bit guest. Memory overhead (memory consumed in host because of guest running) is substantially higher with 64-bit though.
If you have running linux x86_64 should be faster because the backward-compatibility for i386 is not necessary
Same if you have a optimized x86-System with binarys for i686 instead of i386
But most mainstream-distributions are comatible with i386
And why isn't VMI Paravirtualization supported for 64-bit guest? Is it just that it isn't finished in RC1 or is it not technically possible? How much of a gain does this give in real world scenarios?
It's not necessary, because architectural changes in 64-bit guest kernels and increased capabilities available to our hypervisor when running in long mode (i.e. 64-bit mode) mean that it's not worth it - guests run fast enough without the need for additional paravirtualization.
Speaking personally (not for VMware, or our hypervisor team), I'm in agreement with Avi Kivity (the lead KVM developer) on this topic. Ignore the inflammatory headline and read the post:
has anyone any benchmarks on this? Will the usual performance gain (usually slim) that you get when running on real hardware apply when you go from 32 to 64 bit guests? Or is there some obscure penalty for running 64-bit guest?
Forgot to respond to this part:
- There's no obscure penalty.
- I believe that a 64-bit guest will run faster than its 32-bit counterpart in benchmarks and most real-world scenarios, but if you care about performance, you should never rely on people's opinions and you should benchmark the workloads you care about to make sure that your assumptions hold.
Thank u very much, interesting reading!