[Pipet Devel] node-to-node communication: you can have it 3 ways!

jarl van katwijk jarl at casema.net
Thu Mar 23 01:57:20 EST 2000

> >
> > What does overflow-overflow comm. mean:
> > 1) overflow NODE <-> NODE communication
> > or 2) overflow SUBNET <-> SUBNET communication?
> Definitely 2).  I'm talking about instances of Overflow or Overflow sessions
> (same applies to GMS).

Good, Jean OK too?

> Loci was not going to dictate how applications communicate, if they have their
> own protocol.  This has a practical implication for the computational
> biologists in our group, because many Internet-based databases are on the Web
> (and therefore use HTTP), and many are developing CORBA API's.  If we (the
> collaborative) then want to make such a database a node in our system, should
> the database's own API be translated?  Why not let it work the way it is
> designed to, and let VSH (or whatever) just 'broker' the connection?

Yes, that's why GMS is based upon RAW data transfer (sequencial shunks of 4kb).
Let have the applications interprent the data, VSH should orientate on

> Now, regarding Overflow-to-Overflow communication, if we consider an
> independently running Overflow system (not used as a library) to be an
> application that can communicate with another instance of itself, why not let
> it?

Because it's too unpredictable: it's very easy to create a 'overflow' compatible
program that exploits this detail\low level.
Jean, any comments?

> Okay, as for Overflow, you're saying that node-to-node communication should be
> contained within Overflow, like so...
> But I'm saying that if an Overflow node needs to communicate with another
> Overflow node across the Internet, it's not the node that does the
> communicating but Overflow (brokering)...

Ic, but this is where gms is (going to be) powerfull, let the overflow layer
specialise in
processing, let gms do the brokering\security and client\script IO.
It's a question of orientation of the layers. Of the design. Like a
OS kernel: diff. layers
are seperated from each other for a reason, so it's possible to remain the
overview of
the situation and to be able to implement new idears easely.

> or perhaps GMS (non-brokering)...

Yes, they can, only they talk to an application or the gms layer, nerver to an
other subnet.

More information about the Pipet-Devel mailing list