savas parastatidis

Globus to become part of Apache?

I saw this submission/proposal by Globus over at aparche.org. The Globus Alliance is proposing the creation of an Apache subproject for Grid computing with the Globus Toolkit as the basis.

My first feeling was that this could be a good development for the Grid community but then I thought about the issue of standards and wide adoption.

It’s no secret that we here in Newcastle have been advocating for the use of the existing Web Services infrastructure to build internet-scale applications. Of course the Grid application domain requires high-level services (job submission, metadata registries, etc.). We’ve been advocating for such services to be built using basic Web Services technologies. OGSI was the community’s first attempt to build a common infrastructure for all the high-level services. OGSI is now dead but most of its principles have been moved forward and are now part of WS-RF (resource-orientation, exposure of state, lifecycle management of resources, renewal of endpoint references, notification, etc.). The GGF, through the OGSA WG, is promoting the view that WS-RF should be the basis of all the high-level Grid services. This introduces the problem of wide-adoption. With some in the WS community expressing concerns about the approach (see Grek‘s excellent post on “SOAP yes, SOA, no?” as an example) while others have yet to publicly endorse the approach (e.g. Microsoft), I wonder whether the Grid community will have to face more changes in the future for its underlying infrastructure (if WS-RF has to change again to something else). I hope that won’t be the case.

It’s also the issue of standards for the high-level services. Not all of the high-level services have been standardised yet. Is this an area that Apache wants to get involved in? I don’t know. We’ll have to wait and see I guess.

Having said the above, any contribution to an open-source organisation of years of work and source-code development has to be a good thing.

7 responses to “Globus to become part of Apache?”

  1. vangelis
    Very interesting development. I’ve always considered Globus an overrated, huge chunk, of open source software. Is there any chance that they will put some order in it? (Or even write a decent set of manuals)
  2. Chris Smith
    I’d actually argue that having some readily available and well known implementations of specifications would lead to wider scale adoption. You don’t standardize first and implement second. If the implementation is shown to be cumbersome to use, or not efficient, or whatever, then the standards community can rightly move on to something else. So I think this is good for Grid implementations in general. Screw the standards! 🙂
  3. Savas Parastatidis
    Hey Chris, I will have to disagree. If you widely deploy an implementation of a non-standard then you have interoperability issues with other implementations. Unless you want to make the assumption that everyone should deploy your own implementation. You develop a standard through experimentation and not through wide deployment. If we “screw the standards” as you suggest, then we give up our chances for interoperability.
  4. Chris Smith
    Trying to get interoperability through decree (i.e. develop standards then implementations) is backwards as far as I’m concerned. The best standards generally are developed from a number of competing implementations where interoperability is required for bridging gaps. Experimentation only gets you so far, as the stakes for failure are generally low. WS-RF is a great example. The OGSA-WG is pushing this as the guts of OGSA but there are very few (if any) “Grid implementations” which leverage the WS-RF technology. Standards are for interoperability, not for architecture development.
  5. Savas Parastatidis
    Chris, I agree that you can’t design and agree on a standard without implementations first. That’s not what I said in my reply above. I said that we shouldn’t be encouraging the wide deployment of an implementation before we have agreed on standards. Also, I never suggested that standards should be used for architecture development. You come up with the architectural principles first and then you design your specifications which are then developed into standards in order to guarantee interoperability. The Grid folks decides to create a resource-oriented architecture, for which they needed an infrastructure (WS-RF). That’s fair enough. So, working on an architecture and then on implementations of it is one thing. Promoting a particular implementation as a de facto standard without any governing specs in place, that’s a different matter.
  6. Chris Smith
    Fair enough Savas. I understand what you’re saying, and perhaps the creation of an apache project for this WS-RF implementation is an attempt to make it a de facto standard. Ultimately, as somebody who cares more about the potential for developing applications using these technologies, I’m just glad to see more/any implementations.
  7. Savas Parastatidis
    Yup we are in agreement on this. I favour WS-RF implementations because it’s only then we can actually compare approaches. My discussion was about the high-level services though and not WS-RF. Let’s see how things go.