john harney, prashant doshi lsdis lab, dept. of computer science, university of georgia

22
4 th International Conference on Service Oriented Computing Adaptive Web Processes Using Value of Changed Information John Harney, Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Upload: tacy

Post on 19-Mar-2016

24 views

Category:

Documents


0 download

DESCRIPTION

4 th International Conference on Service Oriented Computing Adaptive Web Processes Using Value of Changed Information. John Harney, Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia. Web Process Composition. Supply Chain Process. Preferred Supplier Service - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

4th International Conference on Service Oriented Computing

Adaptive Web Processes Using Value of Changed

InformationJohn Harney, Prashant Doshi

LSDIS Lab, Dept. of Computer Science,University of Georgia

Page 2: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Web Process Composition

Traditional Web process compositions assume static environments

Supply Chain Process

Start Finish Invoke Response

SpotMarket Service

Rate of Order Satisfaction

Preferred Supplier Service

Rate of Order Satisfaction

OtherSupplier ServiceRate of Order Satisfaction

InventoryService

Rate of Order Satisfaction

ResponseInvoke

Page 3: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Web Process CompositionMany environments are dynamic

Supply Chain Process

Start Finish Invoke Response

SpotMarket Service

Rate of Order Satisfaction

Preferred Supplier Service

Rate of Order Satisfaction

OtherSupplier ServiceRate of Order Satisfaction

InventoryService

Rate of Order Satisfaction

ResponseInvoke

Inventorysatisfaction rate

decreases

Preferred Suppliermay be better choice

Page 4: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Optimal Web Process Composition

• Underlying objective– Web process optimality

• Depends on how accurately the environment is captured

• Requires finding any changes that may have occurred

Page 5: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Motivating Scenario – Supply Chain

Page 6: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Motivating Scenario – Supply Chain

• How does process environment change?– Example: Supply Chain (Inventory service)

• Rate of satisfaction of a supplier service – Eg Inventory satisfaction rate decreases or increases

• Cost of using a service– Cost of the Inventory service decreases or increases

• Other parameters (response time, QoS, etc)

Page 7: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Possible Adaptation Approaches

• Do Nothing (Ignore the changes)– Advantages

• Simple• No additional cost or computational

overhead of adaptation– Disadvantages

• Sub-optimal Web process – Web process can do better

Page 8: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Possible Adaptation Solutions• Query a random provider for relevant

information (eg. Inventory)– Advantages

• Up-to-date knowledge of queried service provider• Performs no worse than “do nothing” strategy

– Disadvantages• Querying for information not free • Paying for information that may not be useful

– Information may not change Web process

Page 9: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Overview of Our Approach• VOC – Value of Changed Information

– Decides if obtaining information is:• Useful

– Will it induce a change in optimality of Web process?• Cost-efficient

– Is the information worth the cost of obtaining it?

• Extension of VOI (Value of Information)

Page 10: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Overview of Our Approach• VOC

– Measures how “badly” the current process is performing in changed environment

– Defined as the difference between:• Expected performance of the old process in the

changed environment• Expected performance of the best process in the

changed environment

Page 11: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Web Process Composition Using MDPs• Markov Decision Processes (MDP) (see JWSR 05)

– Definition: M = (S, A, T, C) S : States, A: Actions,

• Actions may be non-deterministic T: Transition function,• States are fully observable S x A (S)

C: Cost function S x A Real

• Perform stochastic optimization using Dynamic Programming

• Value function heuristic :

• Optimal Policy n : S A– (Minimize expected cost)

0)(

})'()',,(),({)(0

'

1min

sV

sVsasTasCsVSs

n

Aa

n

)()( minarg sVs n

Aa

n

Page 12: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Web Process Composition Using MDPsS: Feature-based state space using propositions

– E.g. Mftg. Inventory Availability Yes|No|Unknown

A: WS invocations – E.g. Check Mftg. Inventory Status

Check Preferred Supplier Status

T: An estimate of the “ground truth” probabilities– E.g. T(Mftg. Inventory Avail = Yes | Check Mfg. Inventory

Status, Mftg. Inventory Availability = Unknown) = 0.33

C: Costs may be obtained through costing analysis

Π*: Determines which service to invoke at a particular state

Page 13: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Formalizing VOC• Supply Chain Example

– Querying Transition function T (satisfaction rate of suppliers in supply chain)

– Changed Transition function – T’(.|a,s’)– Current Policy Value - Vπ(s|T’) – Best Policy Value - Vπ*(s|T’)

Page 14: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Formalizing VOC• Actual service parameters are not

known– Must average over all possible revised

parameters

– We use a belief of revised values• Could be learned over time

Page 15: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Manufacturer’s BeliefsExample - Beliefs of Order Satisfaction

Page 16: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Adaptive Web Process Composition

…Prov 1 Prov 2 Prov n

VOC VOC VOC

Keep current policy

Query Provider Re-solve policy

if needed

1. Calculate VOC for each service provider involved in Web process

2. Find provider whose changed parameter induces the greatest change in policy (VOC*)

3. Compare VOC* to cost of querying

VOC* < Cost of Querying

VOC* > Cost of Querying

*

Page 17: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Our Services Oriented Architecture

Page 18: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Empirical Results• Simulated volatile Supply Chain & Patient

Transfer scenarios for:– Do Nothing

• keeping the same process– Query random provider

• Obtaining information from one provider at each state

• Reformulate composition => Resolve policy– VOC

• VOC for determining if query is needed• Reformulate composition if need be

Page 19: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Empirical Results• Measured the average process cost over a range of query cost

values– VOC queries selectively -- query random strategy cost grows at a larger

rate than VOC– VOC performs no worse that the do nothing strategy

Supply Chain Web Process Patient Transfer Web Process

Page 20: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Discussion• Web Process environments are dynamic

– Processes must adapt to changes in environment to remain optimal

– Obtaining the revised information is crucial but may be costly

• VOC approach– Obtains revised information expected to be

useful– Avoids unnecessary queries

Page 21: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Future Work• VOC calculations are computationally

expensive– Knowledge of service parameter guarantees

may be used to eliminate unnecessary VOC calculations

Page 22: John Harney,  Prashant Doshi LSDIS Lab, Dept. of Computer Science, University of Georgia

Thank you

Questions