Reputation-Based Semantic Service Discovery

Reputation-based Semantic Service Discovery – Cardiff UniversitySlide No. 1

Reputation-based Semantic Service Discovery

��

��

��

��

ETNGRID 2004Presented on 14th June 2004

Slide No. 2

��

Motivating ExampleThe Research ProblemThe Traditional ApproachesA better Approach

FrameworkFramework OverviewMatchmaker and Service ComposerReputation Management

ConclusionsFuture Studies


��

� Mr Screen Bean is looking for a reliable Toyota Saloon Car selling

Service.


��

S1 S2 S3

Sells Toyota Cars

Sells Toyota Cars

Sells Saloons Cars


��

� Use of UDDI

UDDI

FindService

ServiceNameCategory

services

<find_service generic="2.0" xmlns="urn:uddi-org:api_v2"><name>ToyotaCarSellingService</name><categoryBag>

<keyedReference tModelKey=“21525-25365-2589-2“keyName=“automobile" keyValue=“car" />

</categoryBag></find_service>

Limitations

S1(ToytaCarService)+

S2S1(ToytaCarService)

?

? UDDI cannot help automatically locate services based on service capabilities and behaviours (i.e. Trust).


��

� Use of Semantic Matchmaking

Matchmaker

FindService

Ontologyservices

Limitations

<profile:Profile rdf:ID="RequestToyotaSellService"><input><profile:ParameterDescription rdf:ID="Price_Input"><profile:parameterName>Price</profile:parameterName"><profile:restrictedTo rdf:resource="Concepts.daml#Price"\></profile:ParameterDescription></input><output><profile:ParameterDescriotion rdf:ID="Car_Output"><profile:parameterName>ToyotaSaloon</profile:parameterName"><profile:restrictedTo rdf:resource="Vehicle.daml#ToyotaSaloon"\></profile:ParameterDescription></output></profile:Profile>

S1(ToytaCarService)+

S2 (ToytaCarService)+

S3(SaloonCarService)

Matchmakers cannot help automatically locate services based on service behaviours (i.e. Trust).

?


��

� A better approach would enable users to:

� Easily and efficiently discovered a reputable service that is more suitable to user’s needs.

� Focus on the conceptual basis of their experiments rather than understanding the low level details of locating services.

� Easy create and share high quality complex workflows.


��

Grid computing efforts adopt Web services technologies, i.e. Web Services Resource Framework. Our approach is relevant for deployment with WSRF.

Slide No. 9

��

Motivating Example� The Research Problem� The Traditional Approaches� A better Approach

FrameworkFramework OverviewMatchmaker and Service ComposerReputation Management

ConclusionsFuture Studies


!��"�#��

Discovery MangerService

MatchmakerService

ComposerService

Reputation MangerService

ServiceRepository Rulebase


��"��

� Compares service request with service advertisements.

� Ensures the reputation metrics of the advertised service meet the

requirements of the request.

� Implementation is based on the Paolucci’s algorithm.

M. Paolucci, T.Kawamura, T.Payne, and K.Sycare. “Semantic matching of Web services capabilities”, Processding of the 1st

International Conference on Semantic Web

Reference


��"��$��%&

Plug-in MatchI/O of Advert

and Request “similar”

Request

AdvertS = (0,1)

class subsumption

Exact MatchI/O of Advert

and Request match

Advert

Request

S=(0,1)Data TypeMatching

Reputation MetricsMatching

+

Paolucci’s algorithm overview


��

� Discovery Manager Service (DMS) requests the Service Composer

(SC) if the Matchmaker is unable to retrieve a service.

� SC puts together combination of services that can provide the

required functionality and match the requested reputation metric.

� CS uses a dynamic adaptive algorithm using two different sources

of information:

�Rule base: CS queries a rule base to retrieve a rule which can

provide a composition template.

�Chaining Services: CS attempts to create a chain of services

that when put together can fulfil the user objective.


��$��%&

� Rule base: CS queries a rule base to retrieve a rule which can

provide a composition template.

� CS attempts to semantically match the inputs and outputs of each

element in the template with services in the repository.

� If matching does not succeed, CS attempts to find another rule that can

decompose the template further (recursively).

� CS will then query the service repository to ascertain if any service

match the rule.

� Services are connected together into a workflow graph based on the

control constructs specified in the rule.


��

� What constitute “good” reputation is a subjective criterion.

� Users may want services that have good reputation rating in

multiple contexts

� Contexts: accessibility, or reliability (or both)

� Three phases are involved in computing the reputation of a service:

1. Reputation Interrogation Phase (RIP).

2. Reputation Rating Phase (RRP).

3. Reputation Computation Phase (RCP).


��$��%&

� Chaining Services: CS attempts to create a chain of services that when put

together can fulfill the user objective.

Algorithm

�For each service available, find a service that matches the output of

the service requested. Let one such service be Sn.

�Ascertain the input of Sn. Find a service that can generate the input

for Sn. Let this service be S(n -1)

�This process is iterated until the input of the service S(n-x) matches

the input of the service requested.

�Create the workflow which specifies the order of execution of the

components S1 to Sn.


� A user requests service reputation from a RMS.

� The reputation request can either be

� A request for the overall reputation score of

a service

� The reputation score of a service within a

particular context

� The aggregation of a set of contexts.

� A user rates a service based on his observations about the service capability.

� The rating is then published to the RMS.

� Relying on the service users to provide feedback to themselves – unlink the P2P reputation mechanisms.

��

� Three Phases are

involved:

� RIP

� RRP

� RCP

� RMS computes the reputation of a service by evaluating several ratings from other users that interacted with the service in the past.


Rating the Availability of a service.

� A user sends a service request to invoke a particular service.

�� '��

service

�The service is off-line.�The request is rejected because of high system workload or a system fault.RMS

�The user sends feedback to the RMS. �The feedback is either 0 or 1


Rating the Reliability of a service.

� A user sends a service request to invoke a particular service.

�� '��

Service Negotiate SLA

SLA

SLA establishedInvoking the servicebased on SLASLA Violation

�SLA violation implies that the service was not executed successfully.RMS

�The user sends feedback to the RMS. �The feedback is one the following values: { -2, -1, 0, 1, 2}


Rating the Reliability of a service (cont..).

� A service user rates service behaviour by examining the terms in the SLA

with his observation during service execution.

� As users cannot monitor the service execution directly, users compute the

estimated execution time test.

�� '��$��%%&

�t = tgen - test

Time Difference

Actual

Execution

Time

Estimated Time


Rating the Reliability of a service (cont..).

� The user evaluates his perception abut the value of t and sends a rating to

RMS.

� Rating must be a natural number between [-2, 2].

�� '��$��%%&

Not very Reliable-1

Unreliable-2

No evaluation0

Reasonable1

Reliable2

MeaningValue


�� '��

� Two types of service are supported:

� Atomic – executed by a single service provider.

� Composite – combined response from multiple providers.

Generating reputation metrics for atomic services

� RMS receives a reputation interrogation about a particular list of services.

� The request message contains the context in which the user is interested.

� The reputation score of a service within a particular context is computed as the

average rating of the ratings:

NcsR N

csrv�=),(

),(


�� '��$��%%&

� The reputation score of a service within multiple contexts is computed as

the weighted sum of the reputation score of each context:

),(*)(1�

=

=n

iii csRsR αReputation of service

s within all contexts

The weight attachedto a particular context

�The weight of each context reflects its importance to a particular set of users.

�Each time a user interrogate the reputation of a service within a particular

context, the weight of that context is increased.


�� '��$��%%&

The Decay Function

� The reputation is associated with a service decays with time.

� A damping function is introduced.

� To compute the decay function R(s,c)new , we evaluate how long ago a

particular rating was generated:

dcsRcsR oldnew

1*),(),( =

Interval between present time and t


�� '��$��%%&

Generating reputation metrics for composite services.

� CS composes services if the MSS is unable to retrieve a matching service.

� The composite service is constructed from several services with different

reputation scores.

� Four different structures to compose services.


�� '��$��%%&

� Four different structure to compose services:

A B C

(a) Sequence Structure (b) loop Structure

A B C

AB

D

(c) Parallel Structure

CA

BD

(d) Condition Structure

C

+

-


�� '��$��%%&

� Four different structure to compose services:

A B C

(a) Sequence Structure (b) loop Structure

A B C

AB

D

(c) Parallel Structure

CA

BD

(d) Condition Structure

C

+

-

Lemma: If the reputation of A within context c is rv(a,c) and the reputation of B within context c is rv(b,c), and the reputation of A is independent of the reputation of B, and the composite service C = A + B is composed as a sequence structure, then the reputation for the composite service C is defined by: rv(a,c) * rv(b,c)

),(0

∏=

N

iicsR

Slide No. 28

��

Motivating Example� The Research Problem� The Traditional Approaches� A better Approach

Framework� Framework Overview� Matchmaker and Service Composer� Reputation Management

Conclusions� Future Studies


! ��"

� The content of the SLA.

� Trusted Monitoring Service. (Third Party).

� Identify the relationship between the reputation and QoS.

� Implementation of the future approach.


��

�� !��"��#��$��

Reputation-Based Semantic Service Discovery

Documents