xc_domain_getfullinfo() gone

xc_domain_getfullinfo() gone

[Andrew Theurer]

I noticed this was gone from libxc.  Would there be any objection to 
adding xc_domain_get_vcpu_info?  I am interested in querying the 
cpu_time for each vcpu for a utility that does something like:

vm-stat

cpu[ util] domN-vcpuM[util]...domY-vcpuZ[util]
------------ --------------------------------------
cpu0[075.4] dom0-vcpu0[000.3] dom1-vcpu1[075.1]
cpu1[083.7] dom1-vcpu2[083.7]
cpu2[069.2] dom1-vcpu3[069.2]
cpu3[075.9] dom1-vcpu0[075.9]
                                                    < time interval>
cpu0[100.0] dom0-vcpu0[000.5] dom1-vcpu1[099.5]
cpu1[099.8] dom1-vcpu2[099.8]
cpu2[099.8] dom1-vcpu3[099.8]
cpu3[099.8] dom1-vcpu0[099.8]

cpu0[100.0] dom0-vcpu0[000.3] dom1-vcpu1[099.7]
cpu1[099.7] dom1-vcpu2[099.7]
cpu2[099.7] dom1-vcpu3[099.7]
cpu3[099.7] dom1-vcpu0[099.7]

cpu0[100.0] dom0-vcpu0[000.6] dom1-vcpu1[099.4]
cpu1[099.7] dom1-vcpu2[099.7]
cpu2[099.7] dom1-vcpu3[099.7]
cpu3[101.4] dom1-vcpu0[101.4]

And while we're on this subject, I wanted to track, per phys cpu, 
exec_domain context switches, and store this as ctx_switches in 
schedule_data struct.  I believe tracking context switches would be a 
good stat to have, for example, to expose problems like high domU 
traffic networking on one cpu system.  Any objection to this or suggestions?

Thanks,

-Andrew

Re: xc_domain_getfullinfo() gone

[Kip Macy]

It is already there. It hasn't been exported to python as the main
purpose is to query register state. You're welcome to add it.


                        -Kip




On 5/13/05, Andrew Theurer <habanero@us.ibm.com> wrote:
> I noticed this was gone from libxc.  Would there be any objection to
> adding xc_domain_get_vcpu_info?  I am interested in querying the
> cpu_time for each vcpu for a utility that does something like:
> 
> vm-stat
> 
> cpu[ util] domN-vcpuM[util]...domY-vcpuZ[util]
> ------------ --------------------------------------
> cpu0[075.4] dom0-vcpu0[000.3] dom1-vcpu1[075.1]
> cpu1[083.7] dom1-vcpu2[083.7]
> cpu2[069.2] dom1-vcpu3[069.2]
> cpu3[075.9] dom1-vcpu0[075.9]
>                                                     < time interval>
> cpu0[100.0] dom0-vcpu0[000.5] dom1-vcpu1[099.5]
> cpu1[099.8] dom1-vcpu2[099.8]
> cpu2[099.8] dom1-vcpu3[099.8]
> cpu3[099.8] dom1-vcpu0[099.8]
> 
> cpu0[100.0] dom0-vcpu0[000.3] dom1-vcpu1[099.7]
> cpu1[099.7] dom1-vcpu2[099.7]
> cpu2[099.7] dom1-vcpu3[099.7]
> cpu3[099.7] dom1-vcpu0[099.7]
> 
> cpu0[100.0] dom0-vcpu0[000.6] dom1-vcpu1[099.4]
> cpu1[099.7] dom1-vcpu2[099.7]
> cpu2[099.7] dom1-vcpu3[099.7]
> cpu3[101.4] dom1-vcpu0[101.4]
> 
> And while we're on this subject, I wanted to track, per phys cpu,
> exec_domain context switches, and store this as ctx_switches in
> schedule_data struct.  I believe tracking context switches would be a
> good stat to have, for example, to expose problems like high domU
> traffic networking on one cpu system.  Any objection to this or suggestions?
> 
> Thanks,
> 
> -Andrew
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
>

Re: xc_domain_getfullinfo() gone

[Kip Macy]

Never mind, I was thinking get_vcpu_context, but per-cpu time is
already available in get_vcpu_context.

                          -Kip

On 5/13/05, Andrew Theurer <habanero@us.ibm.com> wrote:
> I noticed this was gone from libxc.  Would there be any objection to
> adding xc_domain_get_vcpu_info?  I am interested in querying the
> cpu_time for each vcpu for a utility that does something like:
> 
> vm-stat
> 
> cpu[ util] domN-vcpuM[util]...domY-vcpuZ[util]
> ------------ --------------------------------------
> cpu0[075.4] dom0-vcpu0[000.3] dom1-vcpu1[075.1]
> cpu1[083.7] dom1-vcpu2[083.7]
> cpu2[069.2] dom1-vcpu3[069.2]
> cpu3[075.9] dom1-vcpu0[075.9]
>                                                     < time interval>
> cpu0[100.0] dom0-vcpu0[000.5] dom1-vcpu1[099.5]
> cpu1[099.8] dom1-vcpu2[099.8]
> cpu2[099.8] dom1-vcpu3[099.8]
> cpu3[099.8] dom1-vcpu0[099.8]
> 
> cpu0[100.0] dom0-vcpu0[000.3] dom1-vcpu1[099.7]
> cpu1[099.7] dom1-vcpu2[099.7]
> cpu2[099.7] dom1-vcpu3[099.7]
> cpu3[099.7] dom1-vcpu0[099.7]
> 
> cpu0[100.0] dom0-vcpu0[000.6] dom1-vcpu1[099.4]
> cpu1[099.7] dom1-vcpu2[099.7]
> cpu2[099.7] dom1-vcpu3[099.7]
> cpu3[101.4] dom1-vcpu0[101.4]
> 
> And while we're on this subject, I wanted to track, per phys cpu,
> exec_domain context switches, and store this as ctx_switches in
> schedule_data struct.  I believe tracking context switches would be a
> good stat to have, for example, to expose problems like high domU
> traffic networking on one cpu system.  Any objection to this or suggestions?
> 
> Thanks,
> 
> -Andrew
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
>

RE: xc_domain_getfullinfo() gone

[Ian Pratt]

 
> And while we're on this subject, I wanted to track, per phys 
> cpu, exec_domain context switches, and store this as 
> ctx_switches in schedule_data struct.  I believe tracking 
> context switches would be a good stat to have, for example, 
> to expose problems like high domU traffic networking on one 
> cpu system.  Any objection to this or suggestions?

context switches per CPU woul be a useful thing to have -- we already
record this in the s/w perf counters.

Ian