Automated service composition -- your opinion ?

Hello,

I'm PhD student at the University of Modena, Italy.

I'm currently studying the problem of automated service composition,
which -- as I discovered -- is of your domain, at least when looking at 
WS Challenge or SWS Challenge site. Perhaps
I'm meeting the same doubts you had when you were digging into this
domain. Please treat them rather as my doubts, not offences.

It would be great if I you could share with me with your experience or
impressions. If no, please suggest me some better place (person/mailing
list) where I could get comments about my thoughts.


1. This is quite artificial problem, because:

a) It is desired feature of SOA to have automatically discovered
    services which can be later automatically composed.
    But this is hardly possible to verify whether your ideas of
    composition algorithm are effective, because there are no real
    test collections [1], similary like there is TREC collection for
    testing IR techiniques in XML domain. However, TREC collection is
    based on gigabates of real existing data.

b) Some people can say this is a "chicken-egg" problem. There are not
    so much services on the Web, because techniques for describing them,
    discovering and composing are still inmature. But you could also say
    there is no such techniques, because there is not enough web services
    on the Web, and what can be concluded from this -- there is no
    industry request to have them.

c) As a signal from industry then we can perceive cease of supporting
    business UDDI registries of big players such as Microsoft, IBM, SAP
    in January 2006 [2]. There are still Web services (mainly information
    sources), but I can see the tendency, when major players like Amazon
    or IBM provide their own Web services (information sources, cloud
    computing). So there is still few discoverable servicies of similar
    functionality to make automated and adaptive composition reasonable.

d) Web Service Challenge [3] and Semantic Web Service Challenge [4] were
    aiming in creating sequential composition on the base of input/output
    matching (even when enriched with ontology-based subsumption relation
    between parameters). However the idea of service composition is
    something more then matching interfaces, more composition operators
    can be used like parallel execution, loops etc. [5]

2. The idea of automatic composition is kind of 'utopia'.

a) The SOA is a step forward but still goes along the idea of
    *re-usability*.
    When I create my one-machine application, I'm looking for some
    libraries which I can re-use to avoid re-inventing a wheel. However
    I'm not assuming in advance that the library will have the API
    with particular method because searching would result in lots of
    false negatives. This is because I could not assume the architect of
    desired library structured his solution in the way I would. And I'm
    not talking only about interfaces, but also about matching
    pre-conditions, effects, invariants describing functionality of
    a component (like in [5]). The similar methodology would
    problematic when applied in the SOA world.

b) This why, as I could see, people are trying to relax conditions and
    find services which are rather *similar* then *exact* ones;  and
    later on
    they put some hand-made glue (mediators, connectors) among discovered
    services. This why general idea of making steadily evolving service
    taxonomies of UDDI registries [6] leaves some room of flexibility.

3. I can see some similarities to Data Integration domain, where people
    were very enthusiastic about making fully automated systems for
    discovering 100% sure mappings between data structures (e.g. tables
    from two RDBs) but then they shifted their focus on semi-automatic
    mapping: proposing efficient/effective heuristics, leggible GUI
    for presentation mapping suggestions etc. Is it the similar tendency
    in service composition ? Are there e.g. BPEL composition tools that
    suggests which services match to each other and to our requirements ?
    Of course with relational data it is easier, because there's lots of
    them on the net.

So, what's going on in this domain ?? What are the problems to be worth
of research focus and make small steps in the direction of sth real ?

Thank you for your time,

Maciej Gawinecki
---------------------
http://www.ibspan.waw.pl/~gawinec

1. "Evaluation of Semantic Service Discovery -- A Survey and Directions
     for Future Research"
    http://hnsp.inf-bb.uni-jena.de/DIANE/docs/2007_ECOWS_WEWST_slides.pdf
2. http://www.infoworld.com/article/05/12/16/HNuddishut_1.html
3. Web Service Challenge, http://www.ws-challenge.org
4. Main Page - SWS Challenge Wiki, http://sws-challenge.org
5. Contract-based Web Service Composition
    http://www.edoc.hu-berlin.de/docviews/abstract.php?lang=ger&id=27321
6. The Tao of e-business services
    http://www.ibm.com/developerworks/library/ws-tao/

Received on Thursday, 24 July 2008 23:02:45 UTC