Presentation is loading. Please wait.

Presentation is loading. Please wait.

1 Michael Klein et al., Universität Karlsruhe, Germany Combining Query and Preference – An Approach to Fully Automatize Dynamic Service Binding IEEE International.

Similar presentations


Presentation on theme: "1 Michael Klein et al., Universität Karlsruhe, Germany Combining Query and Preference – An Approach to Fully Automatize Dynamic Service Binding IEEE International."— Presentation transcript:

1 1 Michael Klein et al., Universität Karlsruhe, Germany Combining Query and Preference – An Approach to Fully Automatize Dynamic Service Binding IEEE International Conference on Web Services (ICWS 2004) San Diego, CA, July 6-9, 2004 Michael Klein Institute for Program Structures and Data Organization Universität Karlsruhe, Germany DIANE Project http://www.ipd.uni-karlsruhe.de/DIANE/en Birgitta König-Ries Institute for Computer Science Technical University Munich, Germany

2 2 Michael Klein et al., Universität Karlsruhe, Germany Dynamic Service Binding Application Business Process Service A Service B Service C Service D Agile Networks  change inefficient or unavailable service providers on runtime  Enhance efficiency and robustness

3 3 Michael Klein et al., Universität Karlsruhe, Germany State of the Art  How is this typically achieved? Application Service B Description of the desired service Description of the offered service 1.Attach service description to application and services 2.In case of a call, a matcher calculates similarity between request and all offers 3.Best offer is chosen and invoked Matcher [0..1] invoke

4 4 Michael Klein et al., Universität Karlsruhe, Germany Difficulties EASY:  If descriptions are exactly equal: Matcher returns 1.0  If descriptions are obviously different: Matcher returns 0.0 DIFFICULT:  In the common intermediate situations:  The offer differs somewhat from the request  What value in (0,1) is appropriate?

5 5 Michael Klein et al., Universität Karlsruhe, Germany Example for a Service Request :Service :Profile:Booked :CinemaTicket :SeatInShow 2004-07-10 spiderman2:Movie presents effect entity validFor visible date cinema Service Request: I want a Service which books me a seat for Spider Man 2 time 20:00 hortonPlaza:Cinema price <= 8.00

6 6 Michael Klein et al., Universität Karlsruhe, Germany Example for a Service Offer :Service :Profile:Booked :CinemaTicket :SeatInShow 2004-07-10 spiderman2:Movie presents effect entity validFor visible date cinema cinerama6:Cinema Service Offer: I can book you a ticket for Spider Man 2, this saturday at 8:15 pm in the Cinerama 6. time 20:15

7 7 Michael Klein et al., Universität Karlsruhe, Germany Is it a Match? The requestor wanted :SeatInShow time 20:00 …but the offer can only book a ticket for 20:15. This is not ok, or is it? So assign 0.9? :SeatInShow cinema hortonPlaza:Cinema The requestor wanted …but the offer can only book a ticket in the Cinerama 6. Is this ok? Say yes, because it‘s in the same city  Assign 0.8? What is more important for the offerer: A good price, a good time, a near cinema?  Simply take the average, or the minimum, or…? :CinemaTicket price <= 8.00 The requestor wanted …but this information is missing in the offer. Return 0.0 and skip the service? Or just assign 0.5? Or 1.0?

8 8 Michael Klein et al., Universität Karlsruhe, Germany Would you rely on such a matcher to select and execute a service automatically for you? automatically = you are not asked again  No!  But: It‘s not the matcher‘s fault  Even best matcher can only guess what is still ok for you and what deviation you are willing to accept

9 9 Michael Klein et al., Universität Karlsruhe, Germany Main Problem Matcher does not know the preference/tolerance of the requestor. Main Problem The Matcher  has to use general deviation heuristics  or is very conservative and only allows exact matchs  biased matching  often provides many or no matching offers  Requestor has to choose manually or blindly rely on it.  Cannot or is not used for automatic service binding.

10 10 Michael Klein et al., Universität Karlsruhe, Germany Our Approach  Tell the matcher exactly what the preferences are Generic Matcher requestoffers - generic - based on heuristics  biased Personal Matcher request creates offers - special - only based on the specified preferences  unbiased FROMTO

11 11 Michael Klein et al., Universität Karlsruhe, Germany Preference-Containing Requests  Request != single instance describing the perfect service  But: Set of suitable services  degrees of membership from (0,1] determines preference for this service

12 12 Michael Klein et al., Universität Karlsruhe, Germany Introduction of Object Sets spiderman2 Movie contains all existing movies represents one single movie Movie represents a set of movies Membership conditions are defined in a structured manner

13 13 Michael Klein et al., Universität Karlsruhe, Germany Defintion of an Object Set – Type Conditions Type Conditions  Limits the type of objects in the set  An object can only be member of the set, if it has exactly the given type Movie Double Contains only Movie objectsContains only double values

14 14 Michael Klein et al., Universität Karlsruhe, Germany Defintion of an Object Set – Direct Conditions Direct Conditions  Tests directly on the name or value of the object  An object can only be member of the set, if it fulfills the condition Movie == spiderman2 Double < 8.00 Contains only one object: spiderman2 Contains only double values that are smaller than 8.00

15 15 Michael Klein et al., Universität Karlsruhe, Germany Defintion of an Object Set – Property Conditions Property Conditions (only complex types)  Tests on attributs of the object  An object can only be member of the set, if it has the given attribut and its value is in the attached set Movie Contains all movies with mainactor danielRadcliffe and a Duration below 3 hours Actor mainActor == danielRadcliffe Duration < PT180M duration

16 16 Michael Klein et al., Universität Karlsruhe, Germany Missing Strategy  Specifies the behavior in case of missing attributes  Possible values: assume_failed (–), ignore ( ), assume_fulfilled (+) Contains tickets that cost $8.00 or less. Tickets without a price specified are in the set too. CinemaTicket Double price <= 8.00

17 17 Michael Klein et al., Universität Karlsruhe, Germany Connecting Strategy  Specifies how the single property conditions are connected  Possible value: expression from „and“ and „or“ of property conditions Contains movies with Daniel Radcliff as well as action movies and thrillers MovieActor mainActor == danielRadcliffe Genre in {action, thriller} genre mainActor or genre

18 18 Michael Klein et al., Universität Karlsruhe, Germany Type Check Strategy  Weakens the type condition  Possible values: =, sub, super, sub[n], super[n] Contains objects of type Movie and of subtypes of Movies Movie sub

19 19 Michael Klein et al., Universität Karlsruhe, Germany Set Definition – Summary direct conditions connection strategy tcs Set prop. condition ms

20 20 Michael Klein et al., Universität Karlsruhe, Germany Set Definition – Fuzzy Extensions direct conditions connection strategy tcs Set prop. condition ms = sub super fsub fsuper … assume_failed ignore assume_fulfilled assume_value[x] and or min max ws ^ … = in <= ~< ~= sim …

21 21 Michael Klein et al., Universität Karlsruhe, Germany Cinema Request (Revisited) Service ProfileBooked CinemaTicket SeatInShowMovie presents effect entity validFor visible date cinema time price Double ~<= 8.00 Date == 2004-07-10 Time ~< 20:00 <= 21:00 Cinema near(hortonPlaza) min(date, time^2, cinema, visible) ==spiderman2

22 22 Michael Klein et al., Universität Karlsruhe, Germany Unbiased Matching  Now: Unbiased Matcher  Matching = Calculating the membership value of the request set  Matcher exactly takes the preferences of the requestor into account  Personal Matcher  Automatically gained matching result is accepted.

23 23 Michael Klein et al., Universität Karlsruhe, Germany Summary  Main Goal of Service Oriented Computing: Dynamic Service Binding  Up to now: Generic Matcher tests similarity between the offer instance and the request as perfect service instance  Problem: Unknown preferences of the requestor in case of deviations  Matcher uses heuristics and is biased  Result is not accepted and must be postprocessed manually  Our Thesis: Tell the matcher exactly what the preferences are  Possible technique: Fuzzy object sets  Matching is test on set membership  unbiased  Result is accepted without further manual processing

24 24 Michael Klein et al., Universität Karlsruhe, Germany Further Work  Difficult for the requestor to adjust the conditions and strategies  Guide user through process  Offer descriptions can contain variables. Is it possible to derive the assignment from the requestor‘s preferences?

25 25 Michael Klein et al., Universität Karlsruhe, Germany Thank you for your attention! DIANE Project http://www.ipd.uni-karlsruhe.de/DIANE/en

26 26 Michael Klein et al., Universität Karlsruhe, Germany APPENDIX

27 27 Michael Klein et al., Universität Karlsruhe, Germany Difference between Offer and Request  Offer description  Provider knows all details  possible to write down single instance  open parts can be specified as variables  Request description  needs a functionality  does not think of a certain service  several services can be suitable  has preferences among these suitable services  Needs special request description language  Possible to exactly define desired services


Download ppt "1 Michael Klein et al., Universität Karlsruhe, Germany Combining Query and Preference – An Approach to Fully Automatize Dynamic Service Binding IEEE International."

Similar presentations


Ads by Google