The end of paravirt_ops?

[Posted September 23, 2009 by corbet]

The paravirt_ops mechanism provides a way for the Linux kernel, when running in a virtualized mode, to hook efficiently into the hypervisor for privileged operations. Over time, processors have grown hardware features aimed at supporting virtualization, but there has still been a performance benefit to implementing some operations through paravirt_ops. That situation would appear to be changing, though.

VMI is a paravirtualization layer for VMWare, built on top of paravirt_ops. Recently, developers at VMWare ran a series of tests and came to an interesting conclusion: with contemporary hardware, using VMI did not improve the performance of guest systems. Indeed, it made things worse. Reasonable hardware virtualization should be available on almost all systems that matter in the near future, so VMWare's developers have decided that VMI no longer makes sense; they are now planning to remove it.

KVM developer Avi Kivity noted that a similar conclusion had been reached in that camp; KVM will be dropping support for some paravirtualized operations in the near future. That leaves two other systems - Xen and lguest - using paravirt_ops. Xen, it seems, will continue to do so for some time, and lguest is highly unlikely to ever sacrifice sufficient puppies to move to hardware virtualization. So paravirt_ops will remain for a little while yet, but the its eventual demise would appear to be in the cards. When it goes, it may just take lguest with it.

The end of paravirt_ops?

Posted Sep 24, 2009 11:08 UTC (Thu) by lbt (subscriber, #29672) [Link] (2 responses)

Seems an odd story...

Paraphrased:
"We have 4 well known VM services using a kernel capability but new hardware doesn't benefit from it and because 1 OSS and 1 commercial solution aren't that interested in old hardware support we expect the kernel to get rid of it and sacrifice the other 2 OSS solutions even though there may be lots of people using them."

Is there something to do with the fact that the Xen/lguest guys won't support paravirt_ops so it's stagnating (I thought the Xen people were doing a good job after an extended hiatus). Is lguest stagnant?

I don't understand why this feature would this be removed if lguest/Xen still relied on it.

(Yes, I use Xen to run VMs on an oldish (3 yr!!) server with no hw virt support.)

The end of paravirt_ops?

Posted Sep 24, 2009 11:29 UTC (Thu) by liljencrantz (guest, #28458) [Link]

I belive the time frame for dropping paravirt_ops is many years, so it would seem premature to start worrying just yet. But the overall direction seems to be that more and more hardware gets good virtualization support - it seems likely that in a few years, a regular cell phone will be using virualization to allow non-trusted applications to run with decent security, much the same way the PS3 does today. At such a time, I think it's likely that paravirt_ops will not be missed by many.

The end of paravirt_ops?

Posted Sep 24, 2009 12:11 UTC (Thu) by arjan (subscriber, #36785) [Link]

it's more a case of
1) removing parvirt hooks that only these 2 used (reduces complexity and perf cost)
2) critically looking at future requests for paravirt hooks if it turns out that paravirt is only a short term solution

The end of paravirt_ops?

Posted Sep 25, 2009 8:56 UTC (Fri) by pcampe (guest, #28223) [Link] (2 responses)

On this article, "A comparison of software and hardware techniques for x86 virtualization", VMWare said that hardware virtualization was not acceptable from a performance point of view; good to see they have completely changed their point of view.

The end of paravirt_ops: vmware change of attitude

Posted Sep 25, 2009 16:03 UTC (Fri) by giraffedata (guest, #1954) [Link]

[In] this article, "A comparison of software and hardware techniques for x86 virtualization", VMWare said that hardware virtualization was not acceptable from a performance point of view; good to see they have completely changed their point of view.

Did they say it was not acceptable? I think they just said it is slower.

And I don't think point of view is what changed. It is a technical paper, and quite objective.

The end of paravirt_ops?

Posted Oct 5, 2009 9:30 UTC (Mon) by robbe (guest, #16131) [Link]

You forgot to give a link (http://www.vmware.com/pdf/asplos235_adams.pdf)
to the paper and the fact that it is three years old.

So the updated story, in a nutshell, is that the first-generation HW
virtualisation (which was state of the art in 2006) is generally slower
than their binary translation technique. Second-generation (EPT, RVI)
features on the other hand do improve performance, are often used by
default, and even a requirement for some VM configurations. Details:

http://www.vmware.com/files/pdf/perf-vsphere-monitor_mode...
http://www.vmware.com/pdf/Perf_ESX_Intel-EPT-eval.pdf
http://www.vmware.com/pdf/RVI_performance.pdf