Hello What performance penalty I should expect from the I/O under domU ? Does it matter to the I/O if the CPU has hardware virtualization ? Thanks -- -------------------------------------------------- Michael Ben-Nes - Internet Consultant and Director. http://www.epoch.co.il - weaving the Net. Cellular: 054-4848113 -------------------------------------------------- _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Its very, very hard to put a ''rule of thumb'' to something like this. A little info about your setup, or setup you are contemplating would help. --Tim On Sat, 2006-08-26 at 13:44 +0300, Michael Ben-Nes wrote:> Hello > > > What performance penalty I should expect from the I/O under domU ? > > Does it matter to the I/O if the CPU has hardware virtualization ? > > > Thanks > >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Saturday 26 August 2006 5:44 am, Michael Ben-Nes wrote:> What performance penalty I should expect from the I/O under domU ? > Does it matter to the I/O if the CPU has hardware virtualization ?no, HVM capabilities let you run non-modified guests, but doesn''t do a thing (yet) for paravirtualized guests. IOW, if you can run something without HVM, having it won''t make a difference. -- Javier _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Tim Post wrote:> Its very, very hard to put a ''rule of thumb'' to something like this. A > little info about your setup, or setup you are contemplating would help. >I don''t see how the Hard disk / RAM type can change the performance penalty. can you explain why for example sata will have more penalty then SCSI ? The system im building up will be a server with 6 SCSI 15k RPM disks. 2 will be used as RAID 1 and 4 as RAID 10. It will be probably an IBM server.> --Tim > > On Sat, 2006-08-26 at 13:44 +0300, Michael Ben-Nes wrote: > >> Hello >> >> What performance penalty I should expect from the I/O under domU ? >> Does it matter to the I/O if the CPU has hardware virtualization ? >> >> Thanks >> >> >> > > > >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Javier Guerra wrote:> On Saturday 26 August 2006 5:44 am, Michael Ben-Nes wrote: > >> What performance penalty I should expect from the I/O under domU ? >> Does it matter to the I/O if the CPU has hardware virtualization ? >> > > no, HVM capabilities let you run non-modified guests, but doesn''t do a thing > (yet) for paravirtualized guests. > > IOW, if you can run something without HVM, having it won''t make a difference. >You mean the the HVM will not boost a modified linux kernel ( with xen ) ?> > ------------------------------------------------------------------------ > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Saturday 26 August 2006 2:32 pm, Michael Ben-Nes wrote:> > IOW, if you can run something without HVM, having it won''t make a > > difference. > > You mean the the HVM will not boost a modified linux kernel ( with xen ) ?right. there are some interesting projects to use the new CPU capabilities for drivers and better memory management (IIRC), but these are still very experimental at best (and some are still just in conceptualisation phase) -- Javier _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
On Sat, Aug 26, 2006 at 10:31:53PM +0300, Michael Ben-Nes wrote:> > Tim Post wrote: > > >Its very, very hard to put a ''rule of thumb'' to something like this. A > >little info about your setup, or setup you are contemplating would help. > > > I don''t see how the Hard disk / RAM type can change the performance > penalty. can you explain why for example sata will have more penalty > then SCSI ?That''s not what Tim said at all. There are many factors which make up how your domUs will perform. I could give you the answer "42", and it would be about as good as any answer you''ll ever get, given the information you''ve provided. - Matt _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Hi, i want to setup a hylafax server in a domu. The PCI-Device (AVM- Fritzcard) is usable in the domu: ----- hylafax ~ # lspci 00:0a.0 Network controller: AVM Audiovisuelles MKTG & Computer System GmbH A1 ISDN [Fritz] (rev 02) ----- I builded the fcpci module, but if i tryed to load it an error appears: ----- fcpci: module license ''Proprietary'' taints kernel. fcpci: Unknown symbol force_evtchn_callback ----- has anybody a idea - or better a solution? Frank _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
> What performance penalty I should expect from the I/O under domU ? > > Does it matter to the I/O if the CPU has hardware virtualization ?Me, personally, would be interested in the performance in the case that the DomU accesses a block device on the host though the "most optimized" case - namely if the DomU is using the Xen frontend and the Xen backend of the Dom0 to access the device. There are many questions, like: Does Xen allow the DomU to make multiple requests that are then queued on the Dom0 or even passed to the lower level SCSI/SATA driver of the Dom0? And so on ... _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
One thing I''ve noticed is that if you have 2 VMs on a server and one is running a IO intensive task while the other is CPU intensive, the IO task tends to get starved and performs very poorly. This seems like a scheduling problem, so perhaps the new credit scheduler helps with this, although I haven''t tried. So to further complicate how to measure I/O performance there is the issue of interference with other VMs. On 8/28/06, Sven Köhler <skoehler@upb.de> wrote:> > What performance penalty I should expect from the I/O under domU ? > > > > Does it matter to the I/O if the CPU has hardware virtualization ? > > Me, personally, would be interested in the performance in the case that > the DomU accesses a block device on the host though the "most optimized" > case - namely if the DomU is using the Xen frontend and the Xen backend > of the Dom0 to access the device. > > There are many questions, like: > Does Xen allow the DomU to make multiple requests that are then queued > on the Dom0 or even passed to the lower level SCSI/SATA driver of the > Dom0? And so on ... > > > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users > > >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users