Fast checkpoints with COW VM memory

Fast checkpoints with COW VM memory

[Mike Sun]

Hi --

According to the Xen Roadmap for 2006, there was talk of copy-on-write
functionality that could be developed easily from the
save/restore/migration code to enable features such as fast
checkpoints or even VM forks.  Has there been any progress on such
work?  Our research work would be highly interested in something to
this effect, and could contribute to the effort if it's still in
progress.

Thanks,
Mike Sun

Xen Roadmap 2006
"The save/restore/relocate in Xen provides almost all we need to be
able to take copy-on-write snapshots of VM's memory, to be used for
rollback or to checkpoint long running jobs (if the guest is
communicating with other machines then the wider effect of such a
rollback must be considered). To be able to support checkpoints, we
need to extend the tools to coordinate snapshotting of virtual disks
with taking the execution state snapshot. As well as enabling such
checkpoints to be initiated from the control tools, perhaps providing
the ability to trigger them from within the guest would be useful too.
A natural progression from supporting checkpointing would be to enable
"VM forking". Rather than creating a read-only checkpoint, the VM
effectively becomes cloned, running in a different domain, writing to
a CoW snapshot of the disk."

Re: Fast checkpoints with COW VM memory

[Tim Deegan]

At 14:59 -0500 on 05 Dec (1196866790), Mike Sun wrote:
> According to the Xen Roadmap for 2006, there was talk of copy-on-write
> functionality that could be developed easily from the
> save/restore/migration code to enable features such as fast
> checkpoints or even VM forks.

FSVO "easily". :)  I am working on extending the typed P2M work I did a
while ago to allow pages to be marked copy-on-write, and other similar
memory tricks.  It's still in the planning stage, but I hope to have
some patches for comment fairly soon.

Brendan Cully gave a talk on this area at the last Summit, and I know
there is more work going on at UBC too.

Tim

-- 
Tim Deegan <Tim.Deegan@citrix.com>
Principal Software Engineer, Citrix Systems.
[Company #5334508: XenSource UK Ltd, reg'd c/o EC2Y 5EB, UK.]

Re: Fast checkpoints with COW VM memory

[tgh]

hi
I am interested in the COW VM memory ,and are there some code or patch 
about it ,or still in process ,or in plan? and if it has been 
implemented and integrated in the source tree, then ,what is the way to 
use it ? say , fast checkpoint, or VM fork ?or something else? is it 
available to use ? or how to do it ?

Thanks




Tim Deegan 写道:
> At 14:59 -0500 on 05 Dec (1196866790), Mike Sun wrote:
>   
>> According to the Xen Roadmap for 2006, there was talk of copy-on-write
>> functionality that could be developed easily from the
>> save/restore/migration code to enable features such as fast
>> checkpoints or even VM forks.
>>     
>
> FSVO "easily". :)  I am working on extending the typed P2M work I did a
> while ago to allow pages to be marked copy-on-write, and other similar
> memory tricks.  It's still in the planning stage, but I hope to have
> some patches for comment fairly soon.
>
> Brendan Cully gave a talk on this area at the last Summit, and I know
> there is more work going on at UBC too.
>
> Tim
>
>   

Re: Fast checkpoints with COW VM memory

[Mike Sun]

[-- Attachment #1: Type: text/plain, Size: 1334 bytes --]

Hi --

I'm currently in the process of implementing the COW functionality.
It will probably take me some time, but I'll get back to you when I
have a better idea of what the timeline will be.

Mike

On Jan 15, 2008 10:23 PM, tgh <wwwwww4187@sina.com.cn> wrote:
> hi
> I am interested in the COW VM memory ,and are there some code or patch
> about it ,or still in process ,or in plan? and if it has been
> implemented and integrated in the source tree, then ,what is the way to
> use it ? say , fast checkpoint, or VM fork ?or something else? is it
> available to use ? or how to do it ?
>
> Thanks
>
>
>
>
> Tim Deegan 写道:
>
> > At 14:59 -0500 on 05 Dec (1196866790), Mike Sun wrote:
> >
> >> According to the Xen Roadmap for 2006, there was talk of copy-on-write
> >> functionality that could be developed easily from the
> >> save/restore/migration code to enable features such as fast
> >> checkpoints or even VM forks.
> >>
> >
> > FSVO "easily". :)  I am working on extending the typed P2M work I did a
> > while ago to allow pages to be marked copy-on-write, and other similar
> > memory tricks.  It's still in the planning stage, but I hope to have
> > some patches for comment fairly soon.
> >
> > Brendan Cully gave a talk on this area at the last Summit, and I know
> > there is more work going on at UBC too.
> >
> > Tim

[-- Attachment #2: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: Fast checkpoints with COW VM memory

[Mike Sun]

[-- Attachment #1: Type: text/plain, Size: 2346 bytes --]

Hi --

I'm attempting to implement some sort of COW functionality for use in
fast snapshotting, but have a few questions I'm hoping people can help
out with.

As per typical OS COW implementations, I mark the memory pages I want
COW to read only protection.  When I receive the page protection fault
in the hypervisor page fault handler, is there a way for me to
determine if the protection fault is the result my my COW action as
opposed to a protection fault actually incurred by the guest (i.e.
perhaps the guest OS is implementing its own COW on portions of its
memory)?  The only method I can think of is to keep track of the
protection bits of the pages before I have the hypervisor modify them
to read only, and use this information in the hypervisor's page fault
handler to determine what it should reset the page's protection bits
to be after the page has been copied.

Thanks,
Mike

2008/1/15 Mike Sun <msun@gatech.edu>:
> Hi --
>
> I'm currently in the process of implementing the COW functionality.
> It will probably take me some time, but I'll get back to you when I
> have a better idea of what the timeline will be.
>
> Mike
>
>
> On Jan 15, 2008 10:23 PM, tgh <wwwwww4187@sina.com.cn> wrote:
> > hi
> > I am interested in the COW VM memory ,and are there some code or patch
> > about it ,or still in process ,or in plan? and if it has been
> > implemented and integrated in the source tree, then ,what is the way to
> > use it ? say , fast checkpoint, or VM fork ?or something else? is it
> > available to use ? or how to do it ?
> >
> > Thanks
> >
> >
> >
> >
> > Tim Deegan 写道:
> >
> > > At 14:59 -0500 on 05 Dec (1196866790), Mike Sun wrote:
> > >
> > >> According to the Xen Roadmap for 2006, there was talk of copy-on-write
> > >> functionality that could be developed easily from the
> > >> save/restore/migration code to enable features such as fast
> > >> checkpoints or even VM forks.
> > >>
> > >
> > > FSVO "easily". :)  I am working on extending the typed P2M work I did a
> > > while ago to allow pages to be marked copy-on-write, and other similar
> > > memory tricks.  It's still in the planning stage, but I hope to have
> > > some patches for comment fairly soon.
> > >
> > > Brendan Cully gave a talk on this area at the last Summit, and I know
> > > there is more work going on at UBC too.
> > >
> > > Tim
>

[-- Attachment #2: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: Fast checkpoints with COW VM memory

[Samuel Thibault]

Mike Sun, le Mon 04 Feb 2008 22:14:26 -0500, a écrit :
> is there a way for me to determine if the protection fault is the
> result my my COW action as opposed to a protection fault actually
> incurred by the guest (i.e. perhaps the guest OS is implementing its
> own COW on portions of its memory)?

Not that I'm aware of.

> The only method I can think of is to keep track of the
> protection bits of the pages before I have the hypervisor modify them
> to read only,

Yes.

Samuel