[ogsa-wg] RE: GRIDtoday Edition: Tony Hey: 'Challenging Times for GGF & Standards'

[Hiro Kishimoto]

Kerl's email bounced.

I agree with Karl. Actually we have had almost the same discussion at 
today's OGSA-WG call. We should sort out either (1) message level 
standards or (2) APIs for the higher level middleware.
----
Hiro Kishimoto

Date: Thu, 3 Mar 2005 16:10:31 +0700
From: Karl Czajkowski <karlcz at univa.com>
To: Steven Newhouse <sjn5 at doc.ic.ac.uk>
Subject: Re: [ogsa-wg] RE: GRIDtoday Edition: Tony Hey: 'Challenging  Times for
GGF & Standards'

Sorry, but I think this is an example of the confused discussion we
are seeing.  You are clearly talking about programming environments
with terms like "application code interface" and MPI as a rhetorical
example.  The question "to WSRF or not to WSRF", on the other hand, is
really not a programming question at all.  (To Globus Toolkit, or to
WSRF.NET might be a more reasonable programming question.)

MPI is largely a programming methodology, despite its name... it is an
API (and build environment) standard that let you port application
code from one host to another.  MPI did not in any way give
interoperability between platforms so that your application could run
some tasks here, some tasks there, etc. across multiple vendor
platforms.  Specific implementations of MPI have tried to address
platform heterogeneity over time, but unless I have missed some recent
advance, there is still no runtime interop of MPI implementations.

Most of the Web services standards are targeted at a completely
different area of runtime system interoperation.  The WSRF model and
Microsoft's comparable proposals talk about basic ways to structure
message interchange and more importantly on some protocol-level
reflection mechnanisms to help "find" interop at runtime in a
heterogeneous environment.  The only programming methodology they
imply is an accident of current vendor tools and prototypes!

I find it an interesting question whether OGSA is trying to
standardize a message-passing architecture or whether they are trying
to standardize a programming environment.  I think it is rather
ambitious, and counter to current Internet-inspired strategies, to try
to do both at once.

If OGSA want to define programming environments, I think it is very
appropriate that they define another "abstraction layer", namely a
programming language binding to build or access distributed Grid
systems.  Hopefully you intend to model a message-passing system, or
all bets are off as to how well this binding can map to Web service
technologies in a globally distributed setting.

If OGSA do not want to define programming environments but do have a
message-passing architecture that they would like to specify
"concretely" using Web service methods, I think they should continue
this discussion of how WSRF and the MS proposals relate.  But in doing
so, they should not get distracted by red herrings about what an API
looks like, how portable their code might be, etc.

If OGSA is split on this point, I suggest that the people should split
their discussions and proceed in parallel.  It does seem that there is
this recurring dead-end debate where one person is arguing about
message protocol mechanisms with someone else who is arguing about
application programs and their terminology accidentally collides.


karl


On Mar 02, Steven Newhouse loaded a tape reading:
> >The idea of inventing "yet an other abstraction layer" doesn't sound 
> >very productive.
> 
> Neither does the idea of re-writing my application code interface when
> WS-RF 2.0 comes out, or something else gets adopted by another
> community, or...
> 
> Just imagine where the parallel computing world would still be if we
> where still exposed to machine specific networking stacks instead of
> MPI? Layers of abstraction are (within reason) good.
> 
> Also you are seeing caution in the UK. Remember OGSI...? The UK put in
> considerable effort to adopt the 'betamax' of grid web service. Are we
> now being offered VHS grid web services? Or will there be another
> Betamax type revision?
> 
> I'm finding some of the comments in this thread (not Frank's) are bit
> worrying. Should dissenters really be given a 'good talking to' until
> they come round to the WS-RF point of view? Wow... roll on GGF!
> 
> Steven
> -- 
> ----------------------------------------------------------------
> Dr Steven Newhouse                        Tel:+44 (0)2380 598789
> Deputy Director, Open Middleware Infrastructure Institute (OMII)
> Suite 6005, Faraday Building (B21), Highfield Campus,
> Southampton University, Highfield, Southampton, SO17 1BJ,  UK

-- 
Karl Czajkowski
karlcz at univa.com