Dependable Web Service Compositions usng a Semantic Replication Scheme

1. Dependable Web Service Compositions usng a Semantic Replication Scheme LABORATÓRIO DE SISTEMAS DISTRIBUÍDOS – LASID DEPARTAMENTO DE CIÊNCIA DA COMPUTAÇÃO - DCC UNIVERSIDADE FEDERAL DA BAHIA - UFBA Daniela Barreiro Claro1 Raimundo José de Araújo Macêdo2 SBRC 2008 • Dr. Daniela Barreiro Claro is supported by FAPESB (BOL2071/2006). • Prof. Raimundo José de Araujo Macêdo is supported by FAPESB and CNPQ(Edital Universal).

2. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Outline • Motivation • Dependability requirements for WSC • The SAREK approach • Experimental tests • Performance evaluation • Comparison with related work • Conclusion and future directions Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 2

3. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Motivation • Web services are autonomous applications that can be published, located and invoked over the Internet. • Due to their potential for heterogeneous integration, companies are implementing their business as a Web service format. • However, a single Web service cannot fulfill a user request and need to be combined. Web service Composition Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 3

4. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Motivation • Whereas WS specifications cover dependable issues, no one handles the problem of dependable and automatic Web service compositions. • This raises another problem: a single WS failure, thus the failure of the whole composition. • Availability or continuity of service must be taken into account to apply WSC in critical applications. Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 4

5. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Motivation • A commonly used technique for improving availability is to replicate services. • This work tackles the problem of dependability requirements of WSC, using ontologies to form a set of semantically alike replicas. • We propose a framework SAREK where a failure of a primary service can be masked by the execution of another service semantically compatible. Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 5

6. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Required Properties for WSC • Some kinds of faults should be treated by a fault tolerant mechanism in order to reach the goal of a WSC. • unavailability of a Web service • partially operational WS • Internet disconnections • Some mechanisms have been introduced • FT-SOAP, WS-Reliability, WS-Replication, etc Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 6

7. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Required Properties for WSC • Data Consistency • WSC should guarantee the integrity of data in its execution • Computation Availability • It is not possible to assume that all WS in a composition are reliable. A WSC should guarantee availability without knowing the reliability level of single WS • Scalability • WSC ability to handle a growing amount of WS • Transparency • WS is included and removed from a composition in a transparent way Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 7

8. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach The SAREK approach • System model and assumptions • A set of activities m is the number of activities. • A set of services n is the number of services. • Candidate services is a subset of Service for a specific activity • A composition C is a sequence of activities performed by a set of services • WS are implemented as processes • Channels are assumed to be reliable Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 8

9. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach The SAREK Approach • SAREK is a modified and enhanced version of SPOC1 • SAREK is divided into two modules • The Planner Module • Aims to automatically determine the activities for a given composition • The Executor Module • Aims to execute the composition defined by the Planner • Both modules are replicated using a passive replication mechanism. 1. Claro D.B, Albers P. And Hao, J-K. A framework for automatic composition of RFQ Web services. In IEEE SCW/ WSCA-ICWS, Salt Lake City, USA. 2007 Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 9

10. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 10

11. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach The Planner Module • The Planner is divided into two main phases: Planning and Optimization • The Planning phase • It uses an AI Planning algorithm based on preconditions and effects • Interacts with OPS (an OWL ontology) to discover services • Each WS is described using an OWL-S format • This phase aims to determine which activity belongs to the composition in order to fulfill the user request • It main issue is the set of activities A that can reach the given request Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 11

12. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach The Planner Module • The Optimization phase • Optimizes the combination of WS and activities • The values used to optimize are based on estimated values retrieved from each candidate Web service • It seems a quotation system • Produces a set of semantic similar compositions • We used a genetic algorithm called NSGA-II to get the Pareto optimal solutions, i.e. the compositions Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 12

13. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach The Executor Module • This module executes the composition using the prefix mechanism • In order to provide fault tolerance both a transactional approach and a replication mechanism are applied. • Semantic replication scheme • Transactional Level Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 13

14. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach The Executor Module • Semantic replication scheme • Faulty Web service can be replaced by a semantically similar service (transparency property) • Kind of spatial redundancy, because there is a set of compositions that achieve the same goal. • The prefix approach increases performance • Only the failed partition of the composition is re-executed and the prefix . It saves recovery time supposing that the service s3failed. Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 14

15. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach The Executor Module • Transactional Level • Temporal redundancy mechanism, SAREK tries one more time to recover from a possible transient fault • If the problem persists, and no other semantic similar composition can be replaced, SAREK roll back the previous executed Web service • Using ACID or a compensation technique Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 15

16. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Experimental Tests • Case Study • SAREK was applied to a bidding process for repairing public buildings. • Prototype implementation of SAREK • Java 1.5, Apache Tomcat 5.0, Axis 1.3, Jena API 2.3, OWL-S API 1.1.0 and MySQL Database 4.1 • Experiments were carried out in single computer • Inter motherboard Core Duo, processor T2300 1.66 Ghz and 1Gb of RAM. Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 16

17. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Simulated scenario • 4 activities • supplyWood, supplyConcrete, supplyIron and buildStaircase • Each activity can be performed by 2 candidate services Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 17

18. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Simulated scenario • There is one WSDL operations for execution • String executeWS() • Results were produced by 2 runs • Without failures Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 18

19. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Simulated scenario • With failures (services 5, 3, 6 failed) Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 19

20. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Performance Evaluation • Each experiment was run 400 times for calculating the average time and standard deviation • Two kinds of experiments • A composition is fixed • A composition is randomly chosen Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 20

21. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Performance Evaluation • The overhead caused by faults for an increasing the number of forced Web services failures. Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 21

22. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Related Work • Comparison in the light of the required dependability properties for automatic WSC Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 22

23. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Conclusion and future directions • Challenges for achieving Dependable Web service compositions • Suggesting a set of required dependability properties. • To the best of our knowledge, SAREK is the first framework that provides such fault tolerant guarantees in WSC • Future works • Evaluate the fault tolerant mechanism in real scenarios Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 23

24. Experimental Tests Performance Evaluation Comparison with Related Work Conclusion Outline Motivation Properties for WSC SAREK approach Thank you! Daniela Barreiro Claro dclaro@ufba.br Raimundo Macêdo macedo@ufba.br http://www.lasid.ufba.br LaSiD/DCC/UFBa Daniela Barreiro Claro Raimundo Macêdo Dependable Web Service Compositions using a dclaro@ufba.br macedo@ufba.br Semantic Replication Scheme 24