[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [moca] Locations?
Sanjiva Prasad wrote:
Only a part of the programme envisaged by us about 10 years ago (and
enhanced substantially by others) has been studied -- for instance, only
stop/crash models for failure have been explored, and the general
framework of "unit of resource allocation" deserve a lot more work.
But it has also been used as "unit of migration" (Dist Join).
In Nomadic Pi and Dpi, locations have yet another twist,
but I rather let the invontors comment on this. :-)
Yes, the work of Peter Sewell and his collaborators is significant
and also provides important perspectives on the variety of approaches
(not least being Hennessy-Riely).
In the Distributed Pi of Hennessy and Riely locations are units of
computation, places where processes run. Processes might also migrate
between locations in order to use resources. Communication happens only
within locations. DPi treats channels as global entities that may be
found at some location. However, only the type system reveals this (see
Hennessy and Riely in Information and Computation, or Amadio, Boudol and
Lhoussaine in FSTTCS).
In order to semantically define channels as the resources of locations,
we (Ravara, Matos, Vasconcelos, Martins and Lopes) have been studying a
lexical scoping discipline to DPi ('a la Obliq). The underlying
principle is that channels are associated uniquely to a site at creation
time; once established, the link is fixed throughout computation. This
allows one to see channels as resources of sites. Channels are no longer
global names that extend throughout networks, but are rather physical
resources that are created in the site they are placed at. Channel
naming and bindings were tamed in order to obtain this meaning without
the need of recurring to complex type systems.
In a preliminary paper, presented at FGC2003 (an ICALP workshop)
http://www.di.fc.ul.pt/~vv/papers.html#pi,
we show that the tight correspondence between channel naming and
resource identity has the direct consequence of producing programs whose
behavior does not depend on the sites where they run.
The question now is: isn't this just pi? where is the distribution?
antonio
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The "models for mobility" mailing list mailto:moca@xxxxxxxxxxxxxxx
http://www-sop.inria.fr/mimosa/personnel/Davide.Sangiorgi/moca.html