[AG-TECH] AccessGrid 3: What information is available

[Steve Smith]

> I'd like to remedy any lack of information that's out there, and you'll see efforts in the 
> near term to do so.  While we're in this intermediate state, let us know of questions so 
> we can address them.

Thanks Tom.

I have a couple of concerns with the current state of affairs.  One is
that the current AGTK is an implementation rather than a protocol; i.e.
it relies on a particular version of globus and other packages, and it
is all but impossible to interoperate with it.    I would like a simple
method to integrate other app into an AG session without having to rely
on massive third-party libraries.

Another is that the web-services/globus architecture is a poor match for
the semantics of how we use AG; conceptually the AG is
connection-oriented, but globus/WS is oriented towards transient
connections.

I think these two problems have forced to development of AG to focus on
the AGTK, which should really be the simplest component.  This leads to
the third problem, which is that the core communication tools of the AG
(vic and rat) are being neglected, and need a major overhaul.  A lot of
the complexity and poor usability of the AG comes from these tools and
the necessity of the AGTK to work around their limitations (vic
especially),  There is some good work being done on vic replacement but
the Markus' ANU team, but that only addresses part of the problem, and
may not suit all configurations.  I've done some basic work on cleaning
up the core and hope to be able to dedicate more time to it soon, and
Douglas at UQ is doing some work, but we really need more people on it.
Unfortunately it's not a sexy job.

Anyway, I'm currently writing a paper for the APAC conference about some
alternatives to the current direction.  I believe you're heading over
for that conference so it would be good to talk then.

Cheers,
Steve