By the way, there are other problems with Windows on KVM right now, mainly revolving around timer issues that cause the VM to burn unnecessary CPU when idle. Nothing to do with cloudstack, but just thought I'd mention it.
On Mon, Nov 18, 2013 at 9:45 PM, Marcus Sorensen <shadow...@gmail.com> wrote: > This is really borderline trivial to do, a small tweak to > LibvirtVMDef.java and LibvirtComputingResource.java for windows-based > images if we wanted to set the default to something like 4 cores per > socket. However, the best way to do it would really be to use a > tunable like the coresPerSocket as mentioned. In the mean time, I'd > have no problem putting that in as a stop-gap unless someone else is > willing to do the work of getting it in the service offering > immediately. I've done quite a bit of windows on KVM troubleshooting > and I haven't seen any obvious performance differences between 8 > sockets, 1 core each and 2 sockets, 4 cores each. I have zero time for > the next few days, but if someone doesn't get to it I can add it in > within the next week or so. > > On Mon, Nov 18, 2013 at 8:32 PM, Arnaud Gaillard > <arnaud.gaill...@xtendsys.net> wrote: >> Yes exactly, there is a hard limit on the number of socket Windows >> supports. For instance for a Windows 7 the max is 2 and for Windows Server >> 4 (except for the data center edition that has a higher limit). >> >> With the current implementation, if you set 8 vcpu, Windows will only use 2 >> on a windows 7 VM. It currently means that creating a windows VM with a >> good level of performance with Cloudstack + KVM is not really possible. >> >> We are keen to help with the implementation of this feature for KVM however >> we would like to make sure that nothing else is in the pipe regarding the >> management of vCPU and core in service offering. >> >> I think that this will also be usefull to contol it from Cloudstack for >> VMware, and Xen (I don't know if this is possible with Xen). >> >> >> >> >> On Tue, Nov 19, 2013 at 12:10 AM, Edison Su <edison...@citrix.com> wrote: >> >>> Certain Windows server hard code on how many sockets it supports: >>> http://www.openwebit.com/c/how-to-run-windows-vm-on-more-than-2-cores-under-kvm/ >>> >>> > -----Original Message----- >>> > From: Chiradeep Vittal [mailto:chiradeep.vit...@citrix.com] >>> > Sent: Monday, November 18, 2013 1:53 PM >>> > To: dev@cloudstack.apache.org >>> > Subject: Re: Windows support for KVM >>> > >>> > I don't really understand the issue. >>> > What is the difference between >>> > <vcpu>8</vcpu> >>> > >>> > and >>> > >>> > <vcpu>1</vcpu> >>> > <cpu> >>> > <topology sockets='1' cores='8' threads='1'/> </cpu> >>> > >>> > >>> > Why does Windows see only 4 cores in the first case? Is it because the 8 >>> > cores are split across physical sockets? >>> > >>> > >>> > On 11/18/13 6:55 AM, "Arnaud Gaillard" <arnaud.gaill...@xtendsys.net> >>> > wrote: >>> > >>> > >Hello, >>> > > >>> > >A few days ago I created a new Jira ticket for the support of topology >>> > >in network offering for KVM. This is needed in order to support Windows >>> > >VM in KVM (currently the limitation are such that it is not really >>> > >possible to deploy real Windows VM with this configuration). >>> > > >>> > >The JIRA is >>> > >CLOUDSTACK-5071<https://issues.apache.org/jira/browse/CLOUDSTACK- >>> > 5071> >>> > >and >>> > >is refering to a bug opened before: >>> > >CLOUDSTACK-904<https://issues.apache.org/jira/browse/CLOUDSTACK- >>> > 904> >>> > >. >>> > > >>> > >As I have received no comment on it, I would like to know if the >>> > >support of topology in service offering was considered as a priority, >>> > >and if the impact on the GUI was studied? >>> > > >>> > >Cheers! >>> >>> >> >> >> -- >> *Arnaud Gaillard* >> CTO >> Mobile : +41 78 674 58 95 >> >> <https://docs.google.com/file/d/0B2iGziD9SUPURU0yRjhSX1JhU0k/edit?usp=sharing>
