[cap-talk] Moving running applications

[Jed Donnelley]

On 11/26/2007 5:04 PM, Valerio Bellizzomi wrote:
> On 27/11/2007, at 0.24, David Hopwood wrote:
> 
>> Jed Donnelley wrote:
>>> On 11/21/2007 5:18 PM, Norman Hardy wrote:
> 
>> It isn't strictly necessary for a system to use a pure message
>> passing architecture, for it to provide sufficient abstraction from
>> underlying devices to support migration. The design will probably be
>> cleaner if it does use pure message passing, because otherwise special
>> case treatment of particular kinds of device may be required -- but
>> many VMM implementations seem to have managed to implement this anyway.
> 
> It isn't necessary either to use a VMM to do migration. Look at openMosix,
> which can migrate processes from one node to another, and it isn't a VMM,
> it is a Linux kernel extension.

My argument was similar to what I quoted from:

"A Survey of Process Migration Mechanisms", by J. M. Smith, in ACM OS
Review, pp. 28-40, July 1988.

that ends:

"This observation gives us some insight into the
reasons why port or message based systems (such as
DEMOS/MP and Stanford's V system) implement process
migration more easily than other system designs."

Namely that simplified APIs (e.g. the simple "invoke"
API for object/capabilities) facilitate process
migration.  I believe the object/capability API is
about as simple as they come - and much simpler than
the "API" supported by a VMM (devices, etc.)
or by a Unix OS (e.g. Mosix).

However, I do believe there is substantive value
in moving servers between platforms under VMMs, so
I see the moving of VMs as worthwhile work.  The
moving of running applications also seems to have
some value in the context of a cluster like Mosix.

It seems to me that practical value is driving such
migration implementations more than ease of implementation,
which is as it should be I would say.

--Jed  http://www.webstart.com/jed/