Holding slide prior to starting show

Constraint Oriented Negotiation in Open Information Seeking

Environments for the Grid (CONOISE-G) Project

- V.Deora, W.A. Gray, J.Shao, G. Shercliff, P. J. Stockreisser -

Introduction

• Virtual Organisations

• Lifecycle

• Expectation Based QoS

• Moving Conoise to the grid

• Future Work

Conoise Aims and Objectives

• To develop models and techniques that will support the entire VO lifecycle

• Realising this vision requires the development of:– Autonomous agents to represent the different problem

solving entities. – Sophisticated interaction models that enable autonomous

agents to form and interact within groups – Rich knowledge representation and information inter-change

mechanisms

Group MembersAberdeen

Manage agent commitments and provide an intelligent decision making strategyPolicing and governing of Social Laws

CardiffService Discovery and Quality AssessmentQuality Policy & Service Discovery

SouthamptonNegotiation and Coalition Formation Trust and Reputation

Virtual Organisations (VO)

• Consist of semi-independent autonomous entities.

• Each entity has• A range of problem solving

capabilities

• And a range of resources.

Why create a VO?

• Because the virtual organisation is greater than its parts and there is mutual benefit for all participants.

Entity Behaviour

• The entities co-exist but…

• Can compete against one another in a virtual market place

• Can form VOs to exploit a gap in the market

• And attract custom through advertising the cost and quality of it’s services

VO Sequence

VO and Services

Coalition – Forming a partnership with a competitor

New Service – Forming a partnership with an entity with complementary expertise

VO vs Entity

The collection of independent entities will act as a single conceptual unit in the context of the proposed service

(Co-operating and Co-ordinating Activities)

Each entity retains its individual identity outside of this context. It may break a particular partnership if it decides it is in it’s own best interest.

Service Provider (SP) Decisions

Given a request for a service an SP will have several choices.

Several factors may influence the decision making process

Time Constraints

Prior Commitments

Availability of others

Conoise VO Lifecycle

FORMATION OPERATION DISSOLUTION

RESTRUCTURING

- Elimination

- Expansion

Formation

• An entity attempts to form a VO to meet user requirements (Requester Agent)

• Identifies current services (Yellow Pages Agent)• Issues a call for proposals • Needs to decide which proposals to accept to form VO

(QoS Agent)

Problems…• An RA may wish to break an existing commitment in

order to take part in a more lucrative contract.• When is it most profitable for an agent to initiate a VO?• How will RA deal with such decisions?

Operation/Restructure

• Needs to meet contractual Agreements

• Will have to adapt to market changes– New Service Provider– Decreased Utility

Dissolution

• Why would a VO need to disband?• Why is the organisation not formalised?

– To survive in a rapidly changing market environment (Trends, Competition etc…)

– Cost of formalisation– Limitations imposed by alliance may be too

restrictive to development– Mutual benefit may only be available once

A Example Scenario

Package required by customer:– Movie subscription.– News service.– >50 free text

messages per month.

30 free phone minutes per month.

SP Movies (pcm)

News (daily)

Text (#free)

Phone (#free)

SP1 10 24

SP2 72

SP3 120 30mins

SP4 5 30mins

Conoise Architecture

•Built on JADE Agent Platform using java

•Agents communication in ACL

JADE

MessagingLifecycleWhite &

Yellow Pages

SP RA QA SP SPYP

Container Container Container

QoS – What Is It?• As Functionality?

If one service offers more functionality than others, then it is considered to offer a better quality

• As Conformance?

If a service honours its “claims”, it is considered to offer good quality

• As Reputation?

Users’ perception of a service’s consistency over time

QoS as Conformance

• QoS(Ai) = f (Aia, Ai

d)

A measure of difference between delivered (Aid) and

advertised (Aia) qualities.

• QoS(S) = Σ wi QoS(Ai)

A weighted average of individual qualities

S

A1 A2 An….

Service

QoSAttributes

An Example

• QoS(fr) = f (fra=24, frd=22) = 22/24 = 0.92

• QoS(av) = f (ava=7, avd=7) = 7/7 = 1.00

• QoS(news) = 0.8QoS(fr) + 0.2QoS(av) = 0.94

News to PDA

UpdateFrequency Availability

A Real-World Example

How are they derived?

A Real-World Example (cont’d)

1 out of 5

Rating on a hotel room byMr Fussy

A Real-World Example (cont’d)Rating on a hotel room byMr Easy

5 out of 5

Observations

• What do the ratings mean?

• Current approaches do not differentiate individual users’ expectations on QoS

• Need a more user-centric QoS model

An Expectation Based Model

QoS(Ai) = f (Aia, Ai

d)

User rated quality of Ai

User perceived quality on Ai

QoS(Ai) = <Ru(Ai), Eu(Ai), Pu(Ai)>

User expected quality on Ai

An Example• Suppose that we have 3 providers (SP1, SP2

and SP3) who offer news services to PDA

• We wish to establish their qualities in terms of offering 24 updates per day (Frequency)

• Assume that we have had 6 users (U1, U2, U3, U4, U5 and U6) who have used the services

• We wish to determine which SP is the best

Conventional Approach

Users SP1

Frequency

SP2

Frequency

SP3

Frequency

U1 0.3 0.3

U2 0.8 0.9

U3 0.3 1.0

U4 0.8

U5 0.5 0.1

U6 0.6 0.3

Aggregate rating

0.50 0.67 0.47

Collect ratings only

SP2 is the best

Our Approach

Users SP1

<E(fr), P(fr), R(fr)>

SP2

<E(fr), P(fr), R(fr)>

SP3

<E(fr), P(fr), R(fr)>

U1 <0.9, 0.7, 0.3> <0.7, 0.5, 0.3>

U2 <0.4, 0.4, 0.8> <0.5, 0.5, 0.9>

U3 <0.8, 0.6, 0.3> <0.4, 0.5, 1.0>

U4 <0.6, 0.6, 0.8>

U5 <0.9, 0.7, 0.5> <0.9, 0.5, 0.1>

U6 <0.9, 0.7, 0.6> <0.7, 0.5, 0.3>

Collect expectation, perception and ratings

Our Approach

Users SP1

<E(fr), P(fr), R(fr)>

SP2

<E(fr), P(fr), R(fr)>

SP3

<E(fr), P(fr), R(fr)>

U1 <0.9, 0.7, 0.3> <0.7, 0.5, 0.3>

U2 <0.4, 0.4, 0.8> <0.5, 0.5, 0.9>

U3 <0.8, 0.6, 0.3> <0.4, 0.5, 1.0>

U4 <0.6, 0.6, 0.8>

U5 <0.9, 0.7, 0.5> <0.9, 0.5, 0.1>

U6 <0.9, 0.7, 0.6> <0.7, 0.5, 0.3>

Collect expectation, perception and ratings

You must give your expectation!

If Expectation is 0.8 …

Users SP1

<E(fr), P(fr), R(fr)>

SP2

<E(fr), P(fr), R(fr)>SP3

<E(fr), P(fr), R(fr)>

U1 <0.9, 0.7, 0.3> <0.7, 0.5, 0.3>

U2 <0.4, 0.4, 0.8> <0.5, 0.5, 0.9>

U3 <0.8, 0.6, 0.3> <0.4, 0.5, 1.0>

U4 <0.6, 0.6, 0.8>

U5 <0.9, 0.7, 0.5> <0.9, 0.5, 0.1>

U6 <0.9, 0.7, 0.6> <0.7, 0.5, 0.3>

Assume range = [E-0.1, E+0.1]

SP1 is the best - QoS(SP1) = 0.43

If Expectation is 0.5 …

Users SP1

<E(fr), P(fr), R(fr)>

SP2

<E(fr), P(fr), R(fr)>

SP3

<E(fr), P(fr), R(fr)>

U1 <0.9, 0.7, 0.3> <0.7, 0.5, 0.3>

U2 <0.4, 0.4, 0.8> <0.5, 0.5, 0.9>

U3 <0.8, 0.6, 0.3> <0.4, 0.5, 1.0>

U4 <0.6, 0.6, 0.8>

U5 <0.9, 0.7, 0.5> <0.9, 0.5, 0.1>

U6 <0.9, 0.7, 0.6> <0.7, 0.5, 0.3>

Assume range = [E-0.1, E+0.1]

SP3 is the best - QoS(SP3) = 1.0

If Expectation is unspecified

Users SP1

<E(fr), P(fr), R(fr)>

SP2

<E(fr), P(fr), R(fr)>

SP3

<E(fr), P(fr), R(fr)>

U1 <0.9, 0.7, 0.3> <0.7, 0.5, 0.3>

U2 <0.4, 0.4, 0.8> <0.5, 0.5, 0.9>

U3 <0.8, 0.6, 0.3> <0.4, 0.5, 1.0>

U4 <0.6, 0.6, 0.8>

U5 <0.9, 0.7, 0.5> <0.9, 0.5, 0.1>

U6 <0.9, 0.7, 0.6> <0.7, 0.5, 0.3>

Our approach falls back a conventional one

SP2 is the best - QoS(SP2) = 0.67

The QA’s Architecture

QoS

Calculator

QoS

Calculator

QoS CollectorQoS CollectorThe QA

E request for ratings E

result ratings matching E RPE

RPE

Ratings DB

ExpectationPerception

Rating+

Service Agreement

Service Monitoring

Summary of Our Approach Main Features

Attempt to calculate QoS in context Dynamically aggregate QoS ratings on a case-by-

case basis

Related Work Rating based QoS measurement QoS calculation in marketing research Collaborative filtering QoS taxonomies

Conoise-G Proposal• Trust & Reputation

– In an untrustworthy environment how reliable are the sources from which we obtain information and services?

• Policing– How do we detect when a contract is broken and what reaction do

we take?

• Quality

• Align Conoise services with a Grid-enabled environment– Determine how ontology and resource discovery mechanisms can

employ services offered within the Grid – Determine required Grid structure to support Conoise work

Conoise as a Grid

• Virtual Organisations present in grid

• Difference is in implementation rather than concept

• Research areas within conoise/conoise-G contribute to grid research

JADE

MessagingLifecycleWhite &

Yellow Pages

Container Container Container Container

Globus

Core

GSI

RLS

GRAM

MDS

Possible Implementations

• Grid enable Agent Platform

• Agent Platform itself would make use of grid services

• Requires no reimplementation of agents

Globus

JADE

MessagingLifecycleWhite &

Yellow Pages

SP RA QA SP SPYP

Container Container Container

RA

Container

CA

Core

GSI

RLS

GRAM

MDS

Possible Implementations

• reimplement agents as grid services

• long development period

• requires additional support services

Globus

Core

GSI

RLS

GRAM

MDS

SP RA QA SP SPYP

CA

Possible Implementations

• Provide grid-interface for core agents

• Can make use of grid services

• Possibly use bridge to allow external grid services to interact with conoise agents

Globus

JADE

MessagingLifecycleWhite &

Yellow Pages

SP RA QA SP SPYP

Container Container Container

RA

Container

CA

Core

GSI

RLS

GRAM

MDS

Possible Implementations

• Allows phased implementation

• Both core agents and service providers can make use of grid services

• Eventually remove agent platform

Globus

JADE

MessagingLifecycleWhite &

Yellow Pages

SP RA QA SP SPYP

Container Container Container

RA

Container

CA

Core

GSI

RLS

GRAM

MDS

Problem Issues

• Progression of grid technology– Proposal written 3 years ago

• Integration of Agents and Grid– How will agents interact with external grid

services and software?– How can we utilise evolving grid standards

for core functions such as security and resource discovery?

Future Work @Cardiff

• Continue research into QoS assessment– QoS attribute aggregation– QoS of composite services– Monitoring of QoS degradation

QoS Attribute Aggregation

• Current model values each quality metric as equal

Avail=0.8

Reliab=0.6

Accur=0.4

0.6

FrmRt=0.3

Reliab=0.3

Accur=1.0

~0.5

• Simply averaging the QoS components does not present an accurate QoS value for the service

• Need a model which produces a more meaningful value

SP

SP

QoS of Composite Services • Consider a composite service A,

composed of services B and C

• Can we use the QoS attributes we already have for B and C to meaningfully assess the QoS of A

A

B C

QoS=0.3 QoS=0.9

QoS=??

• Current model assumes all QoS attributes are present

• Current model assumes 100% confidence in QoS values

• Service Providers may actually be composite services

• How do we determine at which point in the chain a QoS degradation is taking place?

• Based on observation can we predict a QoS degradation before it occurs?

QoS Degradation

Conclusions

• Achievements of conoise– CLP for decision making– Negotiation for VO formation– Expectation Based QoS assessment

• Future Work– VO operation and disbandment– Trust & Policing– Grid implementation