I have two problems about CPU Affinity.
1. On boot
We do not have the way of setting CPU Affinity to each VCPU of guest
domain by using a configuration file.
So we have to execute the commands after starting the guest domain.
2. On reboot
We lost the information of setting CPU Affinity that set while the
guest domain is running after rebooting the guest domain.
So we have to set it again.
To solve latter problem, this is one of the patch that I need to apply.
http://lists.xensource.com/archives/html/xen-devel/2007-03/msg01295.html
Thanks,
Saori Fukuta
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel
> 1. On boot > We do not have the way of setting CPU Affinity to each VCPU ofguest> domain by using a configuration file. > So we have to execute the commands after starting the guestdomain. I thought you could by specifying a list of physical CPUs? Perhaps this got broken by the xapi changes?> 2. On reboot > We lost the information of setting CPU Affinity that set while the > guest domain is running after rebooting the guest domain. > So we have to set it again.Affinity information should certainly be persistent. Best, Ian _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> -----Original Message----- > From: xen-devel-bounces@lists.xensource.com > [mailto:xen-devel-bounces@lists.xensource.com] On Behalf Of Ian Pratt > Sent: 18 May 2007 10:40 > To: Saori Fukuta; xen-devel > Subject: RE: [Xen-devel] Problems about CPU Affinity > > > 1. On boot > > We do not have the way of setting CPU Affinity to each VCPU of > guest > > domain by using a configuration file. > > So we have to execute the commands after starting the guest > domain. > > I thought you could by specifying a list of physical CPUs? > Perhaps this > got broken by the xapi changes?No, the problem that the original post talks about, is that you can''t set affinity of individual VCPU''s to individual PCPU''s, e.g: vcpus=2 cpus=2 Works well, but there''s no way to perform: vcpu0=pcpu0 vcpu1=pcpu1 Other than running "vm vcpu-set <domain> <vcpu> <pcpu(s)>" after the domain was started. -- Mats> > > 2. On reboot > > We lost the information of setting CPU Affinity that > set while the > > guest domain is running after rebooting the guest domain. > > So we have to set it again. > > Affinity information should certainly be persistent. > > Best, > Ian > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel > > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Hi, Ian Thank you for your comments. Detailed description is written by her in following mail. http://lists.xensource.com/archives/html/xen-devel/2007-05/msg00416.html She is out now. I reply it instead. Thanks Atsushi SAKAI "Ian Pratt" <Ian.Pratt@cl.cam.ac.uk> wrote:> > 1. On boot > > We do not have the way of setting CPU Affinity to each VCPU of > guest > > domain by using a configuration file. > > So we have to execute the commands after starting the guest > domain. > > I thought you could by specifying a list of physical CPUs? Perhaps this > got broken by the xapi changes?Cannot specify pcpu-vcpu one-to-one on boot time is her problem. I think this is not coming from xapi changes.> > > 2. On reboot > > We lost the information of setting CPU Affinity that set while the > > guest domain is running after rebooting the guest domain. > > So we have to set it again. > > Affinity information should certainly be persistent.I hope so> > Best, > Ian > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> > I thought you could by specifying a list of physical CPUs? > > Perhaps this > > got broken by the xapi changes? > > No, the problem that the original post talks about, is that you can''t > set affinity of individual VCPU''s to individual PCPU''s, e.g: > vcpus=2 > cpus=2AFAIK, you used to be able to specify a list: vcpus=2 cpus = ["2,3,4","5,6,7"] Maybe this patch never went in. AFAIR it was from Ryan. Ian _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
>I have two problems about CPU Affinity. > >1. On boot > We do not have the way of setting CPU Affinity to each VCPU of guest > domain by using a configuration file. > So we have to execute the commands after starting the guest domain. > >2. On reboot > We lost the information of setting CPU Affinity that set while the > guest domain is running after rebooting the guest domain. > So we have to set it again. > >To solve latter problem, this is one of the patch that I need to apply. >http://lists.xensource.com/archives/html/xen-devel/2007-03/msg01295.htmlHi, The patch cannot solve the problem 1, but it can solve the bug#936. I resend it. Could you apply it? Signed-off-by: Masaki Kanno <kanno.masaki@jp.fujitsu.com> Best regards, Kan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
>> > I thought you could by specifying a list of physical CPUs? >> > Perhaps this >> > got broken by the xapi changes? >> >> No, the problem that the original post talks about, is that you can''t >> set affinity of individual VCPU''s to individual PCPU''s, e.g: >> vcpus=2 >> cpus=2 > >AFAIK, you used to be able to specify a list: > vcpus=2 > cpus = ["2,3,4","5,6,7"]Hi, I tried it. It does not work. # xm create cpus=["2,3,4","5,6,7"] vcpus=2 ./PVdomain.1 Using config file "././PVdomain.1". Error: cpus = [2,3,4,5,6,7]: invalid literal for int(): [2 # xm create cpus="["2,3,4","5,6,7"]" vcpus=2 ./PVdomain.1 Using config file "././PVdomain.1". Error: cpus = [2,3,4,5,6,7]: invalid literal for int(): [2 Best regards, Kan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> >>> > I thought you could by specifying a list of physical CPUs? >>> > Perhaps this >>> > got broken by the xapi changes? >>> >>> No, the problem that the original post talks about, is that you can''t >>> set affinity of individual VCPU''s to individual PCPU''s, e.g: >>> vcpus=2 >>> cpus=2 >> >>AFAIK, you used to be able to specify a list: >> vcpus=2 >> cpus = ["2,3,4","5,6,7"] > >Hi, > >I tried it. It does not work. > ># xm create cpus=["2,3,4","5,6,7"] vcpus=2 ./PVdomain.1 >Using config file "././PVdomain.1". >Error: cpus = [2,3,4,5,6,7]: invalid literal for int(): [2 ># xm create cpus="["2,3,4","5,6,7"]" vcpus=2 ./PVdomain.1 >Using config file "././PVdomain.1". >Error: cpus = [2,3,4,5,6,7]: invalid literal for int(): [2 >Hi Ian, Sorry, it was my mistake. I thought that the cpus parameter is able to specify a list currently. Could you tell me why the patch was not applied? Best regards, Kan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
* Ian Pratt <Ian.Pratt@cl.cam.ac.uk> [2007-05-18 05:36]:> > > I thought you could by specifying a list of physical CPUs? > > > Perhaps this > > > got broken by the xapi changes? > > > > No, the problem that the original post talks about, is that you can''t > > set affinity of individual VCPU''s to individual PCPU''s, e.g: > > vcpus=2 > > cpus=2 > > AFAIK, you used to be able to specify a list: > vcpus=2 > cpus = ["2,3,4","5,6,7"] > > Maybe this patch never went in. AFAIR it was from Ryan.I don''t think we ever took the multi cpumask per vcpu patch. At the time, only the part where we apply the same cpumask defined from cpus="" value to each vcpu when in allowing the credit scheduler to balance within the affinity mask. Couple of threads of discussion here: http://lists.xensource.com/archives/html/xen-devel/2006-08/msg00965.html follow up threads: http://lists.xensource.com/archives/html/xen-devel/2006-08/msg00783.html http://lists.xensource.com/archives/html/xen-devel/2006-08/msg00785.html The only changes that is see committed were: changeset: 11111:438ed1c4b391 user: kfraser@localhost.localdomain date: Tue Aug 15 10:56:59 2006 +0100 summary: [XEND] Apply the domain cpumask fully to every vcpu in the domain. -- Ryan Harper Software Engineer; Linux Technology Center IBM Corp., Austin, Tx (512) 838-9253 T/L: 678-9253 ryanh@us.ibm.com _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Hi, Thank you for your comments ! On Fri, 18 May 2007 10:07:46 -0500 Ryan Harper wrote:> * Ian Pratt <Ian.Pratt@cl.cam.ac.uk> [2007-05-18 05:36]: > > > > I thought you could by specifying a list of physical CPUs? > > > > Perhaps this > > > > got broken by the xapi changes? > > > > > > No, the problem that the original post talks about, is that you can''t > > > set affinity of individual VCPU''s to individual PCPU''s, e.g: > > > vcpus=2 > > > cpus=2 > > > > AFAIK, you used to be able to specify a list: > > vcpus=2 > > cpus = ["2,3,4","5,6,7"]Yes, I was talking about the things described above.> > Maybe this patch never went in. AFAIR it was from Ryan. > > I don''t think we ever took the multi cpumask per vcpu patch. At the > time, only the part where we apply the same cpumask defined from cpus="" > value to each vcpu when in allowing the credit scheduler to balance within > the affinity mask. > > Couple of threads of discussion here: > > http://lists.xensource.com/archives/html/xen-devel/2006-08/msg00965.html > > The only changes that is see committed were: > > changeset: 11111:438ed1c4b391 > user: kfraser@localhost.localdomain > date: Tue Aug 15 10:56:59 2006 +0100 > summary: [XEND] Apply the domain cpumask fully to every vcpu in the domain.Therefore, I can not specify such a list now. If everything is OK, could you apply these patches ? Thanks, Saori Fukuta _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel