Last week, Microsoft, IBM, HP, and Intel published a roadmap document about converging their WS specifications in the resource-orientation, events, and systems management space. There has already been some commentary about the roadmap (Ian Foster‘s “The Holy Grail: Industry-Wide System Management Standards at Last?”, Jim‘s “The Emperor’s Second Set of New Clothes is Gone“, and I am sure many more). If you want to build resource-oriented and/or systems management solutions using Web Services technologies, then the roadmap is a good thing (with ‘good thing’ being a relative term given that not anyone agrees with me :-). Finally, the big players in the WS and systems management space are coming to an agreement.
However, what does this mean to the Grid community? I am sure that given the roadmap towards one set of specifications for resource management and eventing for systems management, many will believe that all applications/services in the Grid domain should be build on top of these. As readers of this blog already know, I don’t believe that all large-scale problems can/should be solved using a resource- or object-based approach. The coupling of addressing/naming/identity does not allow for solutions that cross technology boundaries and leads to brittle applications that require distributed lifetime management which doesn’t scale very well. Anyway, this discussion always ends up sounding like a religious argument so I have long stopped saying anything about it within the Grid community.
Nevertheless, there is something to be said about the second point we were trying to make while I was at the University of Newcastle (“WS-GAF: a framework for building Grid applications using Web Services”, “Web Service Grids: an evolutionary approach”) and also the point that Microsoft‘s recent whitepaper tried to make (“An evolutionary approach to realizing the Grid computing vision”): that of interoperability and steady evolution of the architecture through the adoption of well-supported, widely-adopted (especially amongst key stakeholders), well-documented, stable specifications. Since 2001 the GGF has been trying to agree on infrastructure specifications rather than concentrating on the standardization of high-level, functional specifications (OGSI and then WS-RF). Last week Amazon released Amazon S3, a service that is supposed to support the needs of thousands of users, and still their mechanisms for programmatic access were based on specifications that were available to use in 2001 (plain XML over HTTP or SOAP and WSDL). If Amazon can provide such services using stable specifications, why can’t the Grid services be built in a similar way? Isn’t data-access part of the Grid requirements? And there are so many other examples out there in all aspects of the Grid, or any other, application domain. The push for a common underlying infrastructure is delaying progress.
We are going to witness yet another transition within the GGF (moving from WS-RF to WS-whatever-the-name-is-going-to-be). Surely, the continuous transitions cannot be a ‘good thing’.
One response to “Web Services systems management and the Grid”
Your comment about the GGF is both rather sweeping and rather out of date. There is only a small part of the GGF working on infrastructure – arguably only part of one group. Other groups (including significant parts of the OGSA group) are working on higher-level specs. We are simply adopting an approach whereby the main spec is written independently of the underlying infrastructure. This main spec can then be mapped to different infrastructures as required. So people with WS-RF implementations can have a WS-RF mapping and people with plain WS implementations can use that. We also try to maximise the degree of overlap between such mappings.
Beyond that, there is little the GGF per se can do. It is up to implementors to agree; they can’t be forced by a group of people coming together in a working group.