Mailing List Archive

> Now I'm curious about live migration. Is it possible to live migrate an MPI
> application running on a set of nodes to another set of nodes? Has anyone
> tried that?

If you're using MPI over TCP/IP (which I imagine you are) then it should Just
Work (TM). We have tried live migration with MPI applications but you
shouldn't have any problems moving the VMs around with a cluster.

Cheers,
Mark


-------------------------------------------------------
This SF.Net email is sponsored by:
Sybase ASE Linux Express Edition - download now for FREE
LinuxWorld Reader's Choice Award Winner for best database on Linux.
http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel

> If you're using MPI over TCP/IP (which I imagine you are) then it should
> Just Work (TM). We have tried live migration with MPI applications but you
> shouldn't have any problems moving the VMs around with a cluster.

Sorry I meant to say we have *not* tried live migration with MPI applications.

Note to self: read before clicking send!

Cheers,
Mark


-------------------------------------------------------
This SF.Net email is sponsored by:
Sybase ASE Linux Express Edition - download now for FREE
LinuxWorld Reader's Choice Award Winner for best database on Linux.
http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel

Well, after the issue between two xen dom0 domains is solved there is a
new case we don't
understand:

With two physical domains and 4 guest OSs (2 on each physical node) we
get some rare results with ttcp (b=1000000, l = 1000000):

Lets say we have two guest OSs on physical node A, A1 and A2, and two
guest OSs on physical node B, B1 and B2.

Between A1 and B1 I get 110 000 KB/s (which is almost optimal!)
Between A1 and B2 I get 81 0000 KB/s
Between A2 and B1 I get 94 000 KB/s

Do you have any idea why we get less performance in the last two cases?
It doesn't make sense. It cant be
a bottleneck in the network either because of case 1.(?)

Cheers,
Rune


On Nov 2, 2004, at 5:51 PM, Mark A. Williamson wrote:

>> If you're using MPI over TCP/IP (which I imagine you are) then it
>> should
>> Just Work (TM). We have tried live migration with MPI applications
>> but you
>> shouldn't have any problems moving the VMs around with a cluster.
>
> Sorry I meant to say we have *not* tried live migration with MPI
> applications.
>
> Note to self: read before clicking send!
>
> Cheers,
> Mark
>
>
> -------------------------------------------------------
> This SF.Net email is sponsored by:
> Sybase ASE Linux Express Edition - download now for FREE
> LinuxWorld Reader's Choice Award Winner for best database on Linux.
> http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/xen-devel



-------------------------------------------------------
This SF.Net email is sponsored by:
Sybase ASE Linux Express Edition - download now for FREE
LinuxWorld Reader's Choice Award Winner for best database on Linux.
http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel

Among A1, A2, B1, B2, which ones are domain 0 and which are unpriviledged?

- Bin

On Thu, 4 Nov 2004 18:42:34 +0100, Rune Johan Andresen
<runejoha@idi.ntnu.no> wrote:
> Well, after the issue between two xen dom0 domains is solved there is a
> new case we don't
> understand:
>
> With two physical domains and 4 guest OSs (2 on each physical node) we
> get some rare results with ttcp (b=1000000, l = 1000000):
>
> Lets say we have two guest OSs on physical node A, A1 and A2, and two
> guest OSs on physical node B, B1 and B2.
>
> Between A1 and B1 I get 110 000 KB/s (which is almost optimal!)
> Between A1 and B2 I get 81 0000 KB/s
> Between A2 and B1 I get 94 000 KB/s
>
> Do you have any idea why we get less performance in the last two cases?
> It doesn't make sense. It cant be
> a bottleneck in the network either because of case 1.(?)
>
> Cheers,
> Rune
>
>
>
>
> On Nov 2, 2004, at 5:51 PM, Mark A. Williamson wrote:
>
> >> If you're using MPI over TCP/IP (which I imagine you are) then it
> >> should
> >> Just Work (TM). We have tried live migration with MPI applications
> >> but you
> >> shouldn't have any problems moving the VMs around with a cluster.
> >
> > Sorry I meant to say we have *not* tried live migration with MPI
> > applications.
> >
> > Note to self: read before clicking send!
> >
> > Cheers,
> > Mark
> >
> >
> > -------------------------------------------------------
> > This SF.Net email is sponsored by:
> > Sybase ASE Linux Express Edition - download now for FREE
> > LinuxWorld Reader's Choice Award Winner for best database on Linux.
> > http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
> > _______________________________________________
> > Xen-devel mailing list
> > Xen-devel@lists.sourceforge.net
> > https://lists.sourceforge.net/lists/listinfo/xen-devel
>
> -------------------------------------------------------
> This SF.Net email is sponsored by:
> Sybase ASE Linux Express Edition - download now for FREE
> LinuxWorld Reader's Choice Award Winner for best database on Linux.
> http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/xen-devel
>


-------------------------------------------------------
This SF.Net email is sponsored by:
Sybase ASE Linux Express Edition - download now for FREE
LinuxWorld Reader's Choice Award Winner for best database on Linux.
http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel

A and B are domain 0, A1,A2,B1,B2 are the guest OS systems. I think all
are privileged (?), They are
created with the same config file, exception of the file loopback
configuration and ip config.

Cheers,
Rune

On Nov 4, 2004, at 7:41 PM, Bin Ren wrote:

> Among A1, A2, B1, B2, which ones are domain 0 and which are
> unpriviledged?
>
> - Bin
>
> On Thu, 4 Nov 2004 18:42:34 +0100, Rune Johan Andresen
> <runejoha@idi.ntnu.no> wrote:
>> Well, after the issue between two xen dom0 domains is solved there is
>> a
>> new case we don't
>> understand:
>>
>> With two physical domains and 4 guest OSs (2 on each physical node) we
>> get some rare results with ttcp (b=1000000, l = 1000000):
>>
>> Lets say we have two guest OSs on physical node A, A1 and A2, and two
>> guest OSs on physical node B, B1 and B2.
>>
>> Between A1 and B1 I get 110 000 KB/s (which is almost optimal!)
>> Between A1 and B2 I get 81 0000 KB/s
>> Between A2 and B1 I get 94 000 KB/s
>>
>> Do you have any idea why we get less performance in the last two
>> cases?
>> It doesn't make sense. It cant be
>> a bottleneck in the network either because of case 1.(?)
>>
>> Cheers,
>> Rune
>>
>>
>>
>>
>> On Nov 2, 2004, at 5:51 PM, Mark A. Williamson wrote:
>>
>>>> If you're using MPI over TCP/IP (which I imagine you are) then it
>>>> should
>>>> Just Work (TM). We have tried live migration with MPI applications
>>>> but you
>>>> shouldn't have any problems moving the VMs around with a cluster.
>>>
>>> Sorry I meant to say we have *not* tried live migration with MPI
>>> applications.
>>>
>>> Note to self: read before clicking send!
>>>
>>> Cheers,
>>> Mark
>>>
>>>
>>> -------------------------------------------------------
>>> This SF.Net email is sponsored by:
>>> Sybase ASE Linux Express Edition - download now for FREE
>>> LinuxWorld Reader's Choice Award Winner for best database on Linux.
>>> http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
>>> _______________________________________________
>>> Xen-devel mailing list
>>> Xen-devel@lists.sourceforge.net
>>> https://lists.sourceforge.net/lists/listinfo/xen-devel
>>
>> -------------------------------------------------------
>> This SF.Net email is sponsored by:
>> Sybase ASE Linux Express Edition - download now for FREE
>> LinuxWorld Reader's Choice Award Winner for best database on Linux.
>> http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
>> _______________________________________________
>> Xen-devel mailing list
>> Xen-devel@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/xen-devel
>>
>
>
> -------------------------------------------------------
> This SF.Net email is sponsored by:
> Sybase ASE Linux Express Edition - download now for FREE
> LinuxWorld Reader's Choice Award Winner for best database on Linux.
> http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/xen-devel



-------------------------------------------------------
This SF.Net email is sponsored by:
Sybase ASE Linux Express Edition - download now for FREE
LinuxWorld Reader's Choice Award Winner for best database on Linux.
http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel

>
> A and B are domain 0, A1,A2,B1,B2 are the guest OS systems. I think all
> are privileged (?), They are
> created with the same config file, exception of the file loopback
> configuration and ip config.
>
> Cheers,
> Rune

Are you running on SMP systems? I would guess the performance differs
based on whether DOM0 and the unprivileged guest are runnign on the
same CPU or different CPUs...

-- Keir


-------------------------------------------------------
This SF.Net email is sponsored by:
Sybase ASE Linux Express Edition - download now for FREE
LinuxWorld Reader's Choice Award Winner for best database on Linux.
http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel

Yes, I am running on a SMP system with two CPUs. The nodes are running
on different CPUs, but is this a problem? If the two
guest OSs are unprivileged and run on separate CPUs, they should
perform I/O just as good (or bad) anyway, no?

Cheers,
Rune


On Nov 5, 2004, at 5:47 PM, Keir Fraser wrote:

>>
>> A and B are domain 0, A1,A2,B1,B2 are the guest OS systems. I think
>> all
>> are privileged (?), They are
>> created with the same config file, exception of the file loopback
>> configuration and ip config.
>>
>> Cheers,
>> Rune
>
> Are you running on SMP systems? I would guess the performance differs
> based on whether DOM0 and the unprivileged guest are runnign on the
> same CPU or different CPUs...
>
> -- Keir



-------------------------------------------------------
This SF.Net email is sponsored by:
Sybase ASE Linux Express Edition - download now for FREE
LinuxWorld Reader's Choice Award Winner for best database on Linux.
http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel

Unless your workload maxes out one of the CPUs, or doesn't play as
well as it could with the CPU scheduler. In general it's probably wise
to run your dom0 on another CPU, or at least HyperThread, from other
performance-critical VMs.

The situation on a single CPU should be improveable with tweaking,
however.

-- Keir

>
> Yes, I am running on a SMP system with two CPUs. The nodes are running
> on different CPUs, but is this a problem? If the two
> guest OSs are unprivileged and run on separate CPUs, they should
> perform I/O just as good (or bad) anyway, no?
>
> Cheers,
> Rune
>
>
> On Nov 5, 2004, at 5:47 PM, Keir Fraser wrote:
>
> >>
> >> A and B are domain 0, A1,A2,B1,B2 are the guest OS systems. I think
> >> all
> >> are privileged (?), They are
> >> created with the same config file, exception of the file loopback
> >> configuration and ip config.
> >>
> >> Cheers,
> >> Rune
> >
> > Are you running on SMP systems? I would guess the performance differs
> > based on whether DOM0 and the unprivileged guest are runnign on the
> > same CPU or different CPUs...
> >
> > -- Keir
>



-------------------------------------------------------
This SF.Net email is sponsored by:
Sybase ASE Linux Express Edition - download now for FREE
LinuxWorld Reader's Choice Award Winner for best database on Linux.
http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel