An Integrated Approach to Automated Service Discovery ...€¦ · For humans trivial, for a matchmaker not! Main problems: efficiently determine good configuration ensure constraints

1

DIANE

An Integrated Approach to AutomatedService Discovery, Matchmaking and

Composition

Ulrich Küster1, Birgitta König-Ries1, Mirco Stern2 and Michael Klein2

1University Jena, 2University Karlsruhe, Germany

ukuester|[email protected]|[email protected]

WWW07 - Banff, Canada - May 20072

Motivating Example

screws only

shipping only

integrated sale and shipping

"I need 1000 units of screws at

my factory in Karlsruhe"


Motivating Example (cont.)

� For humans trivial, for a matchmaker not!

� Main problems:

� efficiently determine good configuration

� ensure constraints on composition (and configuration)

� shipping departs where screws are purchased

� specific screw packages must adhere to shippers restriction on weight and size

� …

� Need to express such constraints

� Need a matcher thatautomatically composesservices and handlesthose constraints efficiently

screws only

shipping only

integrated sale and shipping

"I need 1000 units of screws at my factory in

Karlsruhe"


Agenda

1. Introduction and Motivation

2. Background information on DSD(service descriptions and matchmaking)

3. Adaptation to matching multiple effects

4. Evaluation and Summary


DIANE Service Descriptions (DSD)

� Goal: Complete and efficient automation of service discovery, matchmaking and invocation

� Layered lightweight ontology language

� concepts from frame languages, DLs, fuzzy logic and modal logic

� Specific elements for service descriptions

� more intuitive modelling

� fine-tuned expressivity

� efficient yet precise matchmaking


Country

== germany

Cityy

city

locatedIn

Characteristics of DSD (1)

� Set-based declarative descriptions

� Vendors sell thousands of articles

� Shipper provide transportationto a multitude of locations

� Offer described as set of possible effects

� Requests accept differingservices with differing preference

� Request described as fuzzyset of acceptable effects

� Semantics: One out of the described set of effects is requested / will be created

Address


Characteristics of DSD (2)

� Configuration by Variables

� Link inputs and outputs of a service with concepts in thedescription

� Offer-In-Variables (configure offers / provide input)

� Offer-Out-Variables (provide output)

Input: German city

Output: Complete address in

that city


An exemplary offer description (excerpts)

upper

offer : Serviceupper.profile

: ServiceProfilepresents

effect

AddressIN,e,1IN,x,1 Address

IN,e,1IN,x,1

fromAddress toAddress

PhysicalEntity

cargo

WeightMeasureIN,e,1IN,x,1

weight

Country

in {germany, switzerland,

austria}

Cityy

city

locatedIn

Country


austria}

city

locatedIn

WeightUnit

== pound

Double

<=50

valunit

Shipped

City


Matching DSD-Descriptions

� Match fuzzy request set r with configurable offer set o

� Compute fuzzy containment valuesubset Є [0, 1] of o in r(How well is the offer contained in the requested effects?)

� configure o such as to maximize subset

� Implementation descends through description graphs, fills variables with optimal values, recursively computes subset for each element, combines subset values according to strategies


Matching Example (excerpts)

unit

upper

request : Service

upper.profile

: ServiceProfile

presents

effect

Shipped

0.3 * shippingTime + 0.7 * price

toAddresscargo

Address

String

== "Mr. Barney Gumble"

name

String

== "320 East 7th Street"

street

String

== "10021"

zipCode

City

== newyork

city

weight

val

Double

== 10

PhysicalEntity

WeightMeasure

WeightUnit

== pound

upper

offer : Service

upper.profile

: ServiceProfile

presents

effect

AddressIN,e,1IN,x,1

toAddress

PhysicalEntity

cargo

WeightMeasureIN,e,1IN,x,1

weight

Country


austria, usa}

city

locatedIn

WeightUnit

== pound

Double

<=50

valunit

Shipped

City

Multiple Effects ???


Agenda






Expressing constraints – intuitive approach

Address

Screw

ShippedOwned

upper

request : Service

upper.profile

: ServiceProfile

Address

presents

cargoentity

effecteffect

location

fromAddress

toAddress

� Intuitive Approach: Point to the same concept

� Breaks tree structure of descriptions

� Poses problems when configuring an offer

� Need to consider all possible combinations of configurations� not efficient


Expressing constraints – Value Propagation

Address

Screw

ShippedOwned

upper

request : Service

upper.profile

: ServiceProfile

Address

presents

effect 2effect 1

location

fromAddress

toAddress

Screw

Address

Screw

defines $xScrew

uses $x

Address

defines $y

Address

uses $y

entity cargo

� Effects are ordered

� Configuration will be performed at the first encounter (locally and no more globally optimized)

� Chosen value will be propagated

� Preprocessing prevents fatal configurations(can be done efficiently)


A Three-Step Matchmaking Algorithm

Coverage 2Coverage 1

Service 3'(300 packages

of screw #93 to

Karlsruhe)

Service 1'(500 packages

of screw #384

in Hamburg)

Service 2'(shipping from

Hamburg to

Karlsruhe)

Plug-In Matcher

(effect - to – effect, ignores variables)

Service

request

descrip-

tion

Service

offer

descrip-

tions

Service Composer

(1. compute effect coverages, 2. restrict variables)

Final Matcher

(effect - to - effect, fills variables)

Service 4

(shipping

between

US-DE)

Service 3

(vendor

with

shipping)

Service 2

(shipping

in DE,

AU, CH)

Coverage 2

Service 3

(vendor

with

shipping)

Coverage 1

Service 1*

(vendor in

DE)

Service 2*

(shipping

in DE)

*) Variables are restricted (e.g.

warehouses outside of operation

range of shipper are excluded)

MV: 0.7MV: 0.9 MV: 0.8

MV: 0.72

Remaining

offers

Service 1

(vendor in

DE, FR)

Service 5

(vendor in

Asia)

Remaining

offers(all combinationscontaining Service4 and 5 can not berestricted properlyin step 2 and are

dismissed)

MV: 0.7

Final result(two readily

configured options,Coverage 1 beingslightly better than

Coverage 2)

A

B

C


A Matching Algorithm for Multiple Effects (1)

� Matching on an effect-to-effect base

� Ignores variables (does not configure offer yet)

� Greatly reduces number of remaining offers (very precise and selective matching)



� Compute all coverages (polynomial complexity)

� Restrict variables to the cut of linked concepts (linear complexity)

� Filter coverages with empty cut (incompatible services)



� Matching on an effect-to-effect base

� Fills variables / configures offer (includes value propagation)

� Computes overall match value


Agenda






Experimental Evaluation of Approach

� Matchmaker fully implemented (Java)

� 11 service offer descriptions at different granularity

� 11 request descriptions with one to four effects

� 40 options to create matching compositions

� Runtime for each match on average less than 1 s

� Runtime dominated by Plug-In-Match

� composition doesn't compromise efficient matching

0

500

1000

1500

2000

2500

1 2 3 4 5 6 7 8 9 10

groups of request-offer-pairs

run

tim

e (

ms)

Pre-Match Plug-In-Match Service Composition Final Match

Composition


Summary

� Approach integrates automated composition into matchmaking

� Main problems:

� proper configuration of offers

� compatibility of used offer configurations

� Addressed problems by

� extending service description language

� adapting matching algorithm

� introducing value propagation semantics

� Implemented and tested approach

� Working solution for considered case

� Matching time not dominated by composition


Ongoing Work

� Better (but still efficient) configuration possible using techniques similar to top-k query answering?

� Other composition cases apart from multiple effects (like service chaining to transform inputs and outputs)

� Objective evaluation and certification by SWS-Challenge initiative (www.sws-challenge.org) / W3C SWS-Testbed Incubator Group (www.w3.org/2005/Incubator/swsc/)


Thank you for your attendance!

Questions?

Ulrich Küster

[email protected]

DIANE project (services in ad hoc networks)

http://hnsp.inf-bb.uni-jena.de/DIANE/

Documents

An Integrated Approach to Automated Service Discovery ...€¦ · For humans trivial, for a matchmaker not! Main problems: efficiently determine good configuration ensure constraints