IBM Research © 2008 IBM Corporation The Myth and Reality of (Automated) Web Services Composition Dr. Biplav Srivastava

IBM Research

© 2008 IBM Corporation

The Myth and Reality of (Automated) Web Services Composition

Dr. Biplav Srivastavahttp://www.research.ibm.com/people/b/biplav/

IBM T. J. Watson Research Center, USA and India Research Lab, India

Major Collaborators: IBM Research Labs: India, Watson, Zurich; Arizona State University; DAGSTHL Seminar 07061;

University of Georgia, AthensASU: April 2008

IBM Research

© 2008 IBM Corporation2 The Myth and Reality of Web Services Composition

Outline

IT Issues Faced by Businesses Today

Web Services Composition – What is it and Why is it Important– Basics– Typical scenarios

A model for understanding different approaches– Suitability of approaches for different scenarios– Examples

An Update on Progress in Automated WSC– Myth: Resolve scale-up and search issues for WSC composition – Reality: Resolve composition set-up issues (at problem set-up or solving phases)

Emerging Trends – Plan reuse and modification in the context of richer, but unstructured, domain models – Planning in the presence of impoverished domain models: model-lite planning

Conclusion

IBM Research


Major IT Issues Faced by Businesses Today

Business-IT alignment– Are my investments in IT supporting my business?

– Can my investment in IT give competitive edge?

Enterprise Application Integration– Integrating across divisions in the same company

– Integrating with suppliers, partners

Collaboration– Perennial, new global dimension

Asset Reuse– Software reuse is perennial

– Documents, methods, even presentations

IBM Research


Background: Web Services Service Broker

Service Provide

r

Service Requester

WSDLSearch with UDDI and get WSDL of match

Invoke using SOAP

Execute BPEL

Return Solution

What is the service representation?

(Advertised) Instances: A service that can be invoked at a physical URL. It is represented by WSDL. Some semantic representations can compete in this space (OWL-S).

Deployed and Running Instances: Not all advertised services may be running at a given execution time.

Type: Collection of services sharing common capabilities (what they do) but differing in how to access them. Semantic representations should capture this.

IBM Research


A Simple Web Service Composition Scenario

S1

S2

S3

Requester

Search with Requirement

S1 -> S3

…

Service Registry

Composition

Module

S1 S3

Execute based on Composition

Centralized v/s Decentralize Orchestration(S1 could have sent output to S3 directly)

IBM Research


The Potential of WSC Depends on What is a Service?

Service As Business Benefit CS Areas (in addition to AI!)

Online, Deployed Applications

Mashups, Collaboration, New Revenue Streams, Data Integration

User Interfaces, Visualization, Databases/ Streaming

IT Systems EAI Metadata Management, Distributed Systems, Messaging/ Networking

Software Components Software Reuse Software Engineering, Databases

Business Processes Business-IT alignment Business Process, Management Metrics

IBM Research


Case Study: Application Integration

IBM Research


Telco Ecosystem

Service/content providers are often 3rd parties

Telco is the intermediary for delivery of services to enterprises/consumers

– Must improve ease-of-use of its software infrastructure

– Must optimize the utilization of its IT infrastructure

Need to adopt standards-based framework

– Use Web services to build end-user services

– Use semantic annotations allowing service functionality to be programmatically composed

3rd PartyProviders

Telco Enterprise

User

User

User

IBM Research


Composed Service: Helpline Automation

Problem Reporting

RegistryUpdate

CallSetup

Help Desk

MessageDelivery

On Site

Problem Classification

Location-basedAgent Selection

Expert LookupAgent Assignment

Desk-based Expert ID Field Expert ID

Problem Ticket

Problem Ticket, Problem Ticket,

Resolution StatusProblem Ticket,

Resolution StatusProblem Ticket,

Customer Interaction

Top-down or bottom-up

Source: A Service Creation Tool Based on End-to-End Composition of Web Services. V. Agarwal et al, WWW 2005

IBM Research


Creation of a new service

Main IssuesScalability of composition solutionLevel of automationModeling domain informationLeverage industry practices

Specify end-user service capability

Select service providers

Design theflow

Deploy theservice

New service

capabilities

New service

providers

Network / environment

changes

• Manual business process integration

• Use tools like WSAD-IE to create flows and business logic

• Deploy using a flow engine (such as MQWF / WBI SF)

Source: A Service Creation Tool Based on End-to-End Composition of Web Services. V. Agarwal et al, WWW 2005

IBM Research


Synthy System Architecture

Abstract Workflow

(Plan)

Domain Ontolog

y

Service Registry

DeployableWorkflow

ServiceSpecificatio

n

Logical Composer

Physical Composer

Execution Environment

Service CreationEnvironment

Key Components

– Service Capabilities Database

• Information about services available in-house as well as with 3rd party providers

– Telecom Ontology

• Domain-specific terminology

– Logical Composer including Planner

• automated aggregation of services via generative planning-based reasoning techniques

– Physical composer

• Instance selection based on end to end QoS specification

Input

– Requirements document for the new service that needs to be composed

Output

– Deployable workflow representing a composite service

Synthy: http://domino.research.ibm.com/comm/research_people.nsf/pages/biplav.Synthy.html

IBM Research


Synthy IDE

An IDE needs to support several views each applicable to a different kind of role– Service Requester

– Service Developer

– Deployment Engineer

– Administrator

Different technological and Interface Requirements

Service Requester

Administrator

Deployment EngineerService Developer

WebServiceRqmt.

ServiceDiscovery

ServiceSelection

ServiceAggregation

ServiceDeploy--ment

CompositeService

OWL-S,WSDLBPEL BPEL

OWL-S

IBM Research


Synthy IDE: Problem Description Lack of Service Composition tools

– tooling available for creation of web services

– existing prototypes handle only part of the problem

– Need for an Integrated Development Environment (IDE) to ease the process of composition, thereby reducing development time and integration efforts

Tooling Challenges– web services are actively running entities that need to be composed together

– new web services may come up or old ones may go down dynamically, leading to much more frequent changes than in traditional software libraries or components

– the tool should be able to work with components in the runtime environment in addition to offline development modules

– has implications on functionality, interface, performance and runtime behavior of the IDE

IBM Research


Case Study: Online Data Aggregation

IBM Research


Travel Reservation Problem

Source: Getting from Here to There:. Interactive Planning and Agent. Execution for Optimizing Travel. José Luis Ambite et al, IAAI 2002

Online information services– Services are data sources; can be modeled as databases which can be queried with no, or controlled, side-effects– Composite service should be responsive but accuracy can be negotiated– Services are heterogeneously owned, hence relatively autonomous in choosing specifications

IBM Research


Source: Getting from Here to There:. Interactive Planning and Agent. Execution for Optimizing Travel. José Luis Ambite et al, IAAI 2002

Workflow Templates

IBM Research


Case Study: Mashup Advisor

IBM Research


MashUp Advisor Summary*

MashupAdvisor exploits a repository of mashups to provide design-time assistance to the user through relevant suggestions as to what outputs can be generated along with the best plans to generate those outputs. The system has two components: an output ranker, which ranks the outputs based on their popularity scores, and a planner, which uses metric planning algorithms and a configurable utility function. The system takes into account popularity and semantic similarity when recommending services and sources.

Main Contributions:– Recommends new outputs to enrich the mashup– Generates better plans by reusing knowledge built by other users– Saves development time by automatically recommending and linking services

Link to demo

Team: Hazem Elmeleegy Anca Ivan, Rama Akkiraju, Richard GoodwinExtended Team: Biplav Srivastava

IBM Research


System Architecture

MashupA

Repository Manager

CatalogueManager Mashup

Repository

Domain Ontology

StatisticsManager

Semantic Matcher

Output Ranker

Planner

Thesaurus

Mashup Editor Server

(Fusion Server)

Partial Mashup

Ranked Output List

Desired Output

Partial Mashup

Minimum Cost Plan

MashupAdvisor

Mashup Editor Client

(Fusion Client)

Internet

IBM Research


A Framework for Understanding WSCE

IBM Research


A Model for WSCE*

An Overall Web Service Composition and Execution view is important in practice

Web ServiceComposition and

Execution

Specification of Requirement

Available Capabilities

[ Templates, Policies ]

ExecutionTrace

Today, it is not clear what are fundamentally different possible types of WSCE approaches and which type to use in a given scenario?

Events

*Services are assumed to be stateless

IBM Research


Basis for WSCE approaches Are composition and execution separable?

– No, Yes

When does composition happen?– Offline, Online

How does composition happen?– Search-based, Template-based

What information is used for composition?– Service types, Service instances published, Services deployed, Templates/ Policies

How are external events handled at runtime (adaptation)?– On-the-fly, gradual

IBM Research


Search-based Composition Execution

Specifications

X={x1,x2,…x}

T={t1,t2,…t}

Interleaved Approach

On-line

Events

Example: ConGolog, Heracles+Theseus

IBM Research


Monolithic Composition

RuntimeSpecifications

I={i1, i2,… i} X={x1,x2,…x}

W={W1,W2,…WL}

T={t1,t2,…t}

FRE

RIWREW

Monolithic Approach

Off-line On-line

Events

≥

Example: SWORD, SHOP-2 based, Petrinet-based, Astro, METEOR-S

IBM Research


Logical Composition

PhysicalComposition

RuntimeSpecifications

C={c1,c2,…c} I={i1, i2,… i}

S={S1,S2,…SK} W={W1,W2,…WL}

T={t1,t2,…t}

FPC FRE

RAW RIWREW

Staged Approach

Off-line On-lineOff-line

X={x1,x2,…x} Events

≥

Example: Synthy, Self-Serv with web communities (but informal modeling)

IBM Research


Template-based Approach: Creation of a Template

T= {t1,t2,…t}

W={W1,W2,…WL}

S={S1,S2,…SK}

Generalize: remove commitments to get templates

Tem

plates

Staged

Monolithic

Interleaved

Traces

Executable Workflow

Abstract Workflow

IBM Research


Usage of a Template

T= {t1,t2,…t}

Tem

platesW={W1,W2,…WL}

S={S1,S2,…SK}

Staged

Monolithic

Interleaved

Traces

Executable Workflow

Abstract Workflow

Add commitments to generate workflow or trace (Assign values to template parameters)

WSCE

Example: Heracles+Theseus, METEOR-S (Semantic templates, other templates),template-based planning

IBM Research


Basis for WSCE approaches Are composition and execution separable?

– No, Yes

When does composition happen?– Offline, Online

How does composition happen?– Search-based, Template-based

What information is used for composition?– Service types, Service instances published, Services deployed, Templates/ Policies

How are external events handled at runtime (adaptation)?– On-the-fly, gradual

Separable? When How What How

Interleaved No Online Search Services deployed On-the-fly

Monolithic Yes Offline Search Services instances published Gradual

Staged Yes Offline Search Service types, Service instances published

Gradual

Template Yes Offline, Online

Template Templates/ policies, Services instances published, deployed

On-the-fly, Gradual

IBM Research


Comparing Approaches

Interleaved Monolithic Staged Template

Composition Effort

O(λ)O(βλ)

Min: O(λ)O(αλ + Mλ) O(Mλ )

Composition Control

NoneLow:

< RIW; FE >

High:

< RAW;RIW; FC; FE >

High:

<template, underlying composition method>

Ability to Handle Composition

FailureNone Low High Low

Adaptation during Execution

High Medium Medium Low to Medium

Information Modeling

Simple (Instances)

Simple (Instances)Elaborate (Types and Instances)

Elaborate (Templates and Instances)

LimitationSearch should be dead-end

free

Always a time-lag between service

information offline v/s online



Search restricted by template – can cause INCOMPLETENESS; Any restriction of the

underlying composition method

Details in: Understanding approaches for web service composition and execution, Vikas Agarwal, Girish Chafle, Sumit Mittal, Biplav Srivastava, ACM COMPUTE 2008

IBM Research


Two Common Web Service Composition and Execution (WSCE) Scenarios

Online information services– Services are data sources; can be modeled as databases which can be queried with no,

or controlled, side-effects

– Composite service should be responsive but accuracy can be negotiated

– Services are heterogeneously owned, hence relatively autonomous in choosing specifications

– Sub-scenarios:• Comparison product review/ shopping sites, Online travel booking• Mash-ups: ad-hoc data services created by users

Enterprise Application Integration– Services are applications; can be modeled as programs with or without side-effects

– Composite service should accurate but responsiveness can be negotiated

– Services are more homogeneously owned (e.g., intranet); hence some control in choosing specifications can be exercised

– Sub-scenarios:• Service creation to connect internal or partner organizations• Scientific workflows: bioinformatics, Geological sciences

IBM Research


Selecting an Approach for Online Scenario Online information services

– Services are data sources; can be modeled as databases which can be queried with controlled side-effects at the time of purchase

– Composite service should be responsive but accuracy can be negotiated– Services are heterogeneously owned, hence relatively autonomous in choosing

specifications


Composition Effort O(λ)O(βλ)


Composition Control NoneLow:

< RIW; FE >

High:

< RAW;RIW; FC; FE >

High:


Composition Failure Resolution

None Low High Low

Adaptation High Medium Medium Low to Medium

Information Modeling Simple (Instances) Simple (Instances)Elaborate (Types and

Instances)Elaborate (Templates and

Instances)

LimitationSearch should be dead-

end free

Always a time-lag between service information offline

v/s online



Search restricted by template – can cause

INCOMPLETENESS; Any restriction of the underlying

composition method

IBM Research


Selecting an Approach for EAI Scenario

Scalability – with number of services Adaptability – to changes Failure Resolution User Interaction – control and supervision important


Composition Effort O(λ)O(βλ)


Composition Control NoneLow:

< RIW; FE >

High:

< RAW;RIW; FC; FE >

High:


Composition Failure Resolution

None Low High Low

Adaptation High Medium Medium Low to Medium

Information Modeling Simple (Instances) Simple (Instances)Elaborate (Types and

Instances)Elaborate (Templates and

Instances)

LimitationSearch should be dead-

end free

Always a time-lag between service information offline

v/s online



Search restricted by template – can cause

INCOMPLETENESS; Any restriction of the underlying

composition method

IBM Research


An Update on Progress in Automated WSC

IBM Research


Tracing Trends by References

Logic/ Constraints/ Planning– Semantic Web Services , McIlraith, S., Son, T.C. and Zeng, H. IEEE Intelligent Systems. Special Issue on the Semantic Web.

16(2):46--53, March/April, 2001. Copyright IEEE, 2001.– SWORD: A Developer Toolkit for Web Service Composition, Shankar R. Ponnekanti and Armando Fox, WWW 2002– Getting from Here to There:. Interactive Planning and Agent. Execution for Optimizing Travel, José Luis Ambite et al, 2002.

Web Service Composition - Current Solutions and Open Problems, B. Srivastava and J. Koehler, 2003

Semantics, Planning, Model Checking– Semi-automatic Composition of Web Services using Semantic Descriptions, E. Sirin, James Hendler and Bijan Parsia,2003

– Planning and Monitoring Web Service Composition, Pistore et al, 2004

– Automated Composition of Semantic Web Services into Executable Processes, P. Traverso and M. Pistore, 2004

Semantics, Planning, Non Functional requirements– A Service Creation Tool Based on End-to-End Composition of Web Services, V. Agarwal et al, 2005

– Planning with Templates, IEEE Intelligent Systems special issue, 2005

Web Service Composition as Planning, Revisited: In Between Background Theories and Initial State Uncertainty, J. Hoffmann, P. Bertoli, M. Pistore, AAAI 2007.

Understanding approaches for web service composition and execution, Vikas Agarwal, Girish Chafle, Sumit Mittal, Biplav Srivastava, ACM COMPUTE 2008

Domain Specific, Adaptation– SewNet - A Framework for Creating Services utilizing Telecom Functionality, WWW 2008

– Dagstuhl Seminar on Autonomous and Adaptive Web Processes, http://www.dagstuhl.de/programm/kalender/semhp/?semnr=07061

IBM Research


Myth – WSC is Planning and Readily Solvable!

Resolve scale-up and search issues, and WSC will be solved with existing, or incrementally enhanced, planners

WSC has much commonality with Planning– Services are Actions

– Side-affects and inputs/ outputs can be modeled as preconditions/ effects

– Use existing or favorite new methods

Many research papers but not many wide-scale systems– Success in generating compositions

– But generation is one thing, execution another• How to prove composition is correct at runtime?• Are middleware available to execute?• Can domain models be built by typical IT professionals?

Anecdote – – Planner4J family of Java planners: Classical, Metric and Contingent planners in three different composition systems

– Never encountered a composition situation where the scalability of the planner was an issue!

– More work on making planner integratable with external systems• Automatic Parameter Turning (AAAI 05)• Analyzing plans (IAAI05)• Validating input domain and problem models (ISWC 2005)• Generating diverse plans (IJCAI 07)• Reachability analysis to identify potentially relevant services from a large repository• See AAAI06 Nectar paper for details

Planner4J: http://domino.research.ibm.com/comm/research_people.nsf/pages/biplav.Planner4J.html

IBM Research


Reality – Why is WSC not Solved as yet?

Modeling domain is hard– Which expert to believe? Companies in monopolistic situations

(e.g. Windows, SAP) have easier time.

– Can domain models be built by typical IT professionals?

– What is the right level of abstractions?

Handling runtime– How to prove composition is correct at runtime?

– Are middleware available to execute?

Tooling

IBM Research


Emerging Trends in Resolving WSC Issues

IBM Research


Business-Process Driven IT

Figure Source: Model-driven Business Process Platforms, David Frankel, SAP

Packaged Middleware: SAP, Oracle, PeopleSoft

Custom-assembly (IBM)

IBM Research


Business-Process Driven SOA

Business Process

SOA Implementation

(Multiple vendors)

IBM Research


SAPBPR

IBMReAL

Tagged Content

Content Sources

Specialize

40

Reuse business processes

Reuse services implementing business processes

– Reuse plans representing composite services

Plan Modification and Reuse

IBM Research


Model-Lite Planning is Planning with incomplete models

..“incomplete” “not enough domain knowledge to verify correctness/optimality”

How incomplete is incomplete?

Missing a couple of preconditions/effects?

Knowing no more than I/O types?

Source: Model-Lite Planning for the Web Age Masses, S. Kambhampati, AAAI07

IBM Research


Challenges in Realizing Model-Lite Planning

1. Planning support for shallow domain models (Helping human planners)

2. Plan creation with approximate domain models(Planners deal with incompleteness)

3. Learning to improve completeness of domain models (Help complete the model)

Source: Model-Lite Planning for the Web Age Masses, S. Kambhampati, AAAI07

IBM Research


Conclusion

IT Issues Faced by Businesses Today

Web Services Composition is very important– Model for looking at WSC– Case Studies

Looked at progress in automated WSC– Myth: Resolve scale-up and search issues for WSC composition – Reality: Resolve composition set-up issues (at problem set-up or solving

phases)

Emerging Trends – Plan reuse and modification in the context of richer, but unstructured, domain

models – Planning in the presence of impoverished domain models: model-lite planning

Future Issue: Adaptation

IBM Research


Additional Material

IBM Research


Web Services Adaptation

IBM Research


What Causes the Change (inputs/ events)

Structural

– Component specific events (reconfiguration) e.g. failures

– Temporal specific events e.g. timeouts, unexpected (w.r.t. protocol) messages

Contractual: contract as list of attributes and possible values

– Contract violations or cancellation

– Request for re-negotiation from providers

Non-functional (QoS) parameter changes

– Maintain efficiency or optimality

– Avoid contract violations

Changes in the business environment

– E.g. New laws, business models, personnel changes, technology changes

IBM Research


What Could the Techniques Change in the Process (outputs/reactions)

Structural

– Spatial/ component: activity/ topology – addition and deletion

– Temporal: change in ordering constraints

Contractual

– Contract as list of attributes and possible values

• Agreement sections: attributes and tolerable values • Separation clauses; re-negotiation clauses

– Differences

• Adaptation: governed by agreement section• Re-negotiation: governed by re-renegotiation section• Separation followed by negotiation: there are be no constraints

– Approach: choose the change provided by the contract and perform

• To what extent can be done automatically?

IBM Research


A Narrow Mapping from Event Types to Possible Reactions Does Not Exist !!

Default for any event

– Ignore

Marginal QoS Changes

– Ignore

– Re-configuration (using policies)

– Contract cancellation, re-negotiate violating contract, re-adjust other contracts

Contractual cancellation

– Ignore

– Re-configuration (using policies)

– Re-negotiate/ re-adjust violating contract, re-adjust/ cancel other contracts

IBM Research


Adaptation Techniques

Ignore

Hard-wired ad-hoc changes– Procedural hacks

– Using policies/ rule-based systems/ Event-Condition-Action

Mediation– Ontology-based (semantics)

• Type mapping in Meteor-S– Behavioral

• Controller synthesis (Karsten Wolf’s work)

Re-configuration – Using policies

– Using LP (Benatallah)

– Using Logic (Berardi, ConGolog)

– Graph transformation (Hyperedge replacement - Ugo Montanari)

Negotiation and Re-negotiation – Game theory

Ooops … ask the human!

IBM Research


Speculation on Complexity of Handling of Events

Trivial[0]

lowhangingfruit [1]

Challenging[3]

Many years away[4]

Impossible[5]

Unknownevents

MarginalQoS differences(A-WSCE)

Ideasexist[2]

Template-based negotiation

New clauses in contract;Renegotiation

Negotiate

Ignoreevents

Structure

Contract

Business

QoS

Contract cancellation

Protocol violationsE.g. controller synthesis

Semantic mediation

IBM Research © 2008 IBM Corporation The Myth and Reality of (Automated) Web Services Composition Dr. Biplav Srivastava

Documents

ibm corporation

ibm research labs

wsc composition reality

composition setup issues

advertised services

collection of services

service representation

composition centralized