All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Is Xen PV really faster/better than Xen HVM?
danielfeng
Member
I doubt.
I'm using XenServer 6.0 now, and have successfully enabled PV mode on all distros installed including CentOS 6.2, Debian 6.0.3, Ubuntu 11.10, Gentoo and of course Arch. Yes all of them are running in PV mode and all of them have xe-guest-utilities installed (and patched if needed) properly.
I did observe performance difference between 32-bit and 64-bit guests (DomU). Usually 64-bit is better.
I tested their performance using UnixBench. However, NONE of the guests got a higher score when in PV mode than when in HVM mode. Actually HVM scores are like double of PV scores.
Yes with HVM you have extra layers and normally it means extra waste of resources. That fact has been mentioned in almost every PV-HVM debate, however I'm thinking maybe nowadays due to better virtualization support at hardware level, the term "PV is better than HVM" has become a myth indeed?
I'm not an expert of Xen/XenServer, so maybe some pros can explain it better? I do see HVM has better performance than PV according to benchmark.
The CPU is E3-1230/E3-1240.
Comments
I'm not quite sure where you heard that, but it all quite depends - paravirtualization may perform better because the guest is aware of the virtualization and compensates for it, but there are drivers to help shorten the gap with hardware virtualization.
It's not surprising that you would get higher scores with a benching tool, because most hardware virtualization is CPU-based (VT-x & AMD-V), IIRC.
Did you test a difference in IO-/network-operations?