On 8/31/07, Raymond Feng <[EMAIL PROTECTED]> wrote:
>
> What's the use case to have one JVM joining multiple SCA domains (even
> theoretically it's possible)? I agree with Simon that one node should only
> belong to one SCA domain but it seems that he also hinted that one JVM can
> host more than one node.
>
> Thanks,
> Raymond
>
> ----- Original Message -----
> From: "Simon Nash" <[EMAIL PROTECTED]>
> To: <[email protected]>
> Sent: Friday, August 31, 2007 1:20 PM
> Subject: Picture on the Tuscany website SCA Java page
>
>
> > The picture on http://incubator.apache.org/tuscany/sca-java.html doesn't
> > look quite right to me.  It seems to show a node that is split between
> > two domains.  I think an SCA node is part of exactly one domain.  It's
> > possible for more than one domain to run in a single JVM, but from an
> > architectural perspective these domains would each instantiate a
> separate
> > node within the same JVM and would not be part of the same node.
> >
> >   Simon
> >
> >
> >
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: [EMAIL PROTECTED]
> > For additional commands, e-mail: [EMAIL PROTECTED]
> >
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [EMAIL PROTECTED]
> For additional commands, e-mail: [EMAIL PROTECTED]
>
> We are maybe using the term Node to mean different things.

For me a node provides runtime resources (1,2 or 3 below) to one ore more
domains. I used the word Node instead of runtime as we already use the term
runtime in the code (ReallySmallRuntime) and the runtime is currently owned
by the domain.

So the resources that a node might represent are

1 - A part of a JVM
Scenario: some application container is making runtime resources available
to our domain applications

2 - A single JVM
Scenario: a stand-alone tuscany program

3 - A collection of JVMs
Scenario: a high performance compute cluster. We may choose to run component
instances across multiple JVMs for performance reasons but the cluster is
not visible to the topology of the domain. As far is it is concerned the
cluster is one node.

This is  distinct from the distributed domain scenario we have implemented
to date where multiple nodes are used to enable the distribution of the
components in the domain.

4. Raymond has raised something else in another thread about "partitions". I
will let Raymond explain that.

We also individually consider that a Node might be associated with one
domain or many domains.

The key point is what we think a node is. From Simon's comment "these
domains would each instantiate a separate node within the same JVM". To make
this work there must be a mechanism within the JVM that allows domains to be
associated with the JVM. For me this is the node but I'm not hooked on this
particular term.

As for scenarios of multiple domains in a JVM.  I imagine a JVM supporting
multiple domains in the same way that an application server can support
multiple unrelated applications. This may sound a bit marginal for nodes
with resources of types 1


Simon

Reply via email to