12 May
2012
12 May
'12
6:38 p.m.
On Sun, 6 May 2012, Pasi Kärkkäinen wrote:
with "fork performance" I assume you're comparing Xen PV to KVM ? Yes, PV has disadvantage (per design) for that workload, since the hypervisor needs to check and verify each new process page table, and that has some performance hit. For good "fork performance" you can use Xen HVM VMs, which will perform well for that workload, and won't have the mentioned performance hit.
I used both PV and HVM VMs. I don't have the details to hand at the moment, but KVM was superior to both. PV drivers where applicable. I have been running KVM for about 15 months now, with 30 VM's on one host and 38 VM's on another. It has been solid; no problems, but unfortunately I had problems with Xen.
Steve