In a VPS environment (a dom0 which hosts some domU used as VPS by some customers, and you don''t know what customers do with the domUs), do you think it is better to use HVM domU or non-HVM domUs? Bye! _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
> In a VPS environment (a dom0 which hosts some domU used as VPS by some > customers, and you don''t know what customers do with the domUs), do > you think it is better to use HVM domU or non-HVM domUs?HVM will give you greater flexibility in terms of what guest OSes you can offer to customers (e.g. non-Linux, maybe older Linux, etc). PV should achieve better per-guest performance, and therefore let you run more doms on a physical server whilst still maintaining decent performance overall. If you offer HVM Linux, you can still use PV drivers in it. I think there might be PV drivers available for Linux 2.4 guests, even. If you really want flexibility (albeit at the cost of added admin work) then you could offer a selection of options, depending on the customer. Cheers, Mark -- Dave: Just a question. What use is a unicyle with no seat? And no pedals! Mark: To answer a question with a question: What use is a skateboard? Dave: Skateboards have wheels. Mark: My wheel has a wheel! _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Fri, Aug 24, 2007 at 04:02:33PM +0100, Mark Williamson wrote:> > In a VPS environment (a dom0 which hosts some domU used as VPS by some > > customers, and you don''t know what customers do with the domUs), do > > you think it is better to use HVM domU or non-HVM domUs? > > HVM will give you greater flexibility in terms of what guest OSes you can > offer to customers (e.g. non-Linux, maybe older Linux, etc). > > PV should achieve better per-guest performance, and therefore let you run moreThis is a suprise to me. I would have thought that one would get better performance from a HVM guest. Why would PV ''achieve better per-guest performance''? Regards Johann -- Johann Spies Telefoon: 021-808 4036 Informasietegnologie, Universiteit van Stellenbosch "But the end of all things is at hand; be ye therefore sober, and watch unto prayer. And above all things have fervent love among yourselves; for love shall cover the multitude of sins." I Peter 4:7,8 _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Johann Spies wrote:> On Fri, Aug 24, 2007 at 04:02:33PM +0100, Mark Williamson wrote: > >>> In a VPS environment (a dom0 which hosts some domU used as VPS by some >>> customers, and you don''t know what customers do with the domUs), do >>> you think it is better to use HVM domU or non-HVM domUs? >>> >> HVM will give you greater flexibility in terms of what guest OSes you can >> offer to customers (e.g. non-Linux, maybe older Linux, etc). >> >> PV should achieve better per-guest performance, and therefore let you run more >> > > This is a suprise to me. I would have thought that one would get > better performance from a HVM guest. Why would PV ''achieve better > per-guest performance''? >There''s a nice nutshell explanation at http://virt.kernelnewbies.org/TechOverview. Basically certain operations are *extremely* expensive to virtualization. Para-virtualized kernels simply don''t use those operations. (I use the word "simply" in a very genteel sort of way, it''s not that simple.) _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Mon, Aug 27, 2007 at 11:07:22AM +0100, Nico Kadel-Garcia wrote:>> better performance from a HVM guest. Why would PV ''achieve better >> per-guest performance''? >> > There''s a nice nutshell explanation at > http://virt.kernelnewbies.org/TechOverview. Basically certain operations > are *extremely* expensive to virtualization. Para-virtualized kernels > simply don''t use those operations. (I use the word "simply" in a very > genteel sort of way, it''s not that simple.) >Thanks. Johann -- Johann Spies Telefoon: 021-808 4036 Informasietegnologie, Universiteit van Stellenbosch "Use hospitality one to another without grudging." I Peter 4:9 _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
> > There''s a nice nutshell explanation at > > http://virt.kernelnewbies.org/TechOverview. Basically certain operations > > are *extremely* expensive to virtualization. Para-virtualized kernels > > simply don''t use those operations. (I use the word "simply" in a very > > genteel sort of way, it''s not that simple.)Paravirt is faster because it avoids trapping and emulating pretty much anything, and allows the guest to explicitly co-operate with the virtualisation process. HVM is a bit more heavyweight: it allows you to run non-PV guests, but doing this implies you''re doing more emulation. The HVM hardware extensions are not inherently slower than using PV. But emulating a whole machine using them is. This is partly because emulation of some machine features is inefficient, and partly because some aspects of virtualisation (e.g. shadow page tables) are not supported as efficiently by current generations of HVM hardware as they could be (and will be in future). The theoretical optimum would be to use PV *and* HVM at once. i.e. use paravirtualisation when it speeds things up or makes them more simple; use emulation using HVM hardware when it cuts down the diff from the native version of the OS, or when it allows optimisations. An example would be that VT has acceleration for context switches. This could, in principle, be used to be used to accelerate context switches when running PV guests but it currently isn''t. An "ideal" solution would use VT to accelerate this when available, even for PV. The approach of combining elements of hardware-supported virtualisation and software-only paravirtualisation is sometimes called "hybrid virtualisation". There''s a paper on it from this years OLS: https://ols2006.108.redhat.com/2007/Reprints/nakajima-Reprint.pdf Cheers, Mark -- Dave: Just a question. What use is a unicyle with no seat? And no pedals! Mark: To answer a question with a question: What use is a skateboard? Dave: Skateboards have wheels. Mark: My wheel has a wheel! _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Thanks for the explanation and reference Mark.
It has has helped me to realise that I am only scratching the surface
in my understanding of virtualization.
Regards
Johann
--
Johann Spies Telefoon: 021-808 4036
Informasietegnologie, Universiteit van Stellenbosch
"For what shall it profit a man, if he shall gain the
whole world, and lose his own soul?" Mark 8:36
_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users