PDA

View Full Version : Full or paravirtualization for Xen?



riddlermarc
18th June 2007, 03:55 PM
Please can someone give me a realistic explanation of the differences?

virt-manager tells you that paravirtualization is faster than full virtualization, I would have thought that full virtualization was faster as it goes straight to the hardware? I have spent a couple of hours surfing the web, including the ubiquitous wikipedia browsing but I am little clearer.

Also, would either full or paravirtualization be faster (by much?) than VMWare Server? Currently I have to have a couple of 2003 Server installs as VMs to do my dev work; I'd like to move to Xen if possible and if it was at least a good bit faster?

I gather you can't run 2003 Server paravirtualized but have to use full virt, utilising the (in my case AMD X2) cpu extensions? In which case, will it be much/any faster than VMWare Server?

Thank you :)

SlowJet
18th June 2007, 04:38 PM
Could you be confusing the speed of VDisks vs. Real Disk partitions?

From the CPU with the VT type option means the cpu does the part that the virt software did before i. e. Hardware is faster than software.

But the Virtual disks are much slower than real partitions f/s. The testing and portability of VDisks is lost but the idea is that for a tesing or devl environment speed is not as critical as when on a multi-user system running the application under load.

Also, with the VT type CPU, you can just use the Normal install software from the guest and don't need a special XENitized version?

See Virtual-Box software for explanations.

SJ

pjfg
18th June 2007, 05:48 PM
Para-virtualisation is indeed the fastest, but requires the guest OS to be modified so that it is aware that it is virtualised, therefore ruling out Windows as a para-virtualized guest.

Under solutions such as VMware the the virtual guest thinks it has exclusive and priviledged access to the hardware (such as ring 0 on x86, whereas it is really running as a user process at ring 3). The result is that all these calls by the vm guest are intercepted by the host OS; it is this time taken by the host that degrades the performance of the guests. The hardware virtualisation support enabled on newer processors speeds up this process but cannot eliminate it entirely.

riddlermarc
21st June 2007, 05:16 PM
Para-virtualisation is indeed the fastest, but requires the guest OS to be modified so that it is aware that it is virtualised, therefore ruling out Windows as a para-virtualized guest.

Under solutions such as VMware the the virtual guest thinks it has exclusive and priviledged access to the hardware (such as ring 0 on x86, whereas it is really running as a user process at ring 3). The result is that all these calls by the vm guest are intercepted by the host OS; it is this time taken by the host that degrades the performance of the guests. The hardware virtualisation support enabled on newer processors speeds up this process but cannot eliminate it entirely.Thanks for your reply, that goes some way to answering the questions I have :)

So in theory, would running Windows under Xen be as fast/faster than VMWare Server (note, not ESX)? I really want to ditch my 2k3 server but I need a certain performance level for development needs..

pjfg
21st June 2007, 05:27 PM
Sorry, I don't have any figures for the performance of Xen vs VMWare and my only experience of Xen is using para-virtualisation. Personally I would lean towards using VMWare server if I were hosting 2k3 guests, simply based on the features and maturity of the product.