Maurice H. ter Beek (ISTI–CNR, Pisa, Italy) Web Service Composition Approaches: From Industrial Standards to Formal Methods Maurice H. ter Beek (ISTI–CNR, Pisa, Italy) Tuesday, May 15 ICIW 2007 joint work with: Antonio Bucchiarone (Nokia Siemens, Lisbon, Portugal) and Stefania Gnesi (ISTI–CNR)

Outline Compare formal methods w.r.t. characteristics Conclusions Background WS composition approaches Syntactic WS composition Semantic WS composition WS composition characteristics Connectivity, correctness and QoS Compare standardization approaches w.r.t. characteristics Formal methods for WS composition Automata, Petri nets and process algebras Compare formal methods w.r.t. characteristics Conclusions

Background Service-Oriented Computing (SOC) Web Services (WSs) An emerging cross-disciplinary paradigm for distributed computing Changes the way in which software applications are designed, architected (SOA), delivered and consumed Web Services (WSs) Autonomous, platform-independent computational elements, possibly managed by different organizations Described, published, discovered, orchestrated and programmed to build networks of collaborating applications, distributed both within and across organizational boundaries We survey and compare WS composition approaches (both industrial and academic)

Syntactic WS Composition Approaches WS Orchestration (like BPEL4WS) Combines available WSs by adding a central coordinator This orchestrator is responsible for invoking and combining WSs WS Choreography (like WS-CDL) No central coordinator Complex tasks defined by conversations of the participating WSs Composition of peer-to-peer interactions among the collaborating WSs

BPEL vs. WS-CDL Both XML-based BPEL: coordination/composition of WSs (WSDL-based) Processes model the flow of WSs by connecting activities that communicate with external WS providers WS-CDL: choreography description of WSs Interactions describe the information exchange by specifying participants, information and channel Exception handling and compensations supported through exception and finalizer work units Contrary to BPEL, WS-CDL describes a global view of the behavior of the message exchanges of all WSs (rather than behavior defined from viewpoint of one WS)

Semantic WS Composition Approaches Aim: the automation of WS discovery, invocation, composition, interoperation and execution monitoring Describe WSs by explicit, machine-understandable semantics Often rely on ontologies to formalize the domain concepts shared among WSs (like OWL-S and WSMO) The Internet is seen as a globally linked database in which web pages are marked with semantic annotations

OWL-S vs. WSMO Both ontology-based OWL-S WSMO Defines a WS ontology with four main elements: service concept, service profile, service model and service grounding No clear distinction between choreography and orchestration WSMO Defines a model to describe semantic WSs with four main elements: ontologies, WSs, goals and mediators Conceptual design in WSMF, annotations in WSML, execution environment WSMX for dynamic discovery/selection/invocation OWL-S more mature in certain aspects (choreography), while WSMO provides a more complete conceptual model

WS Composition Characteristics I Connectivity: Reliability The ability to deliver responses continuously in time The ability to correctly deliver messages between two endpoints Accessibility The percentage of responses per WS request Exception handling/Compensations What happens in case of an error and how to undo the already completed activities The ability to manage compensations of WS invocations (in case of a failure)

WS Composition Characteristics II Correctness: Safety/Liveness Assertions that some bad event never happens in the course of a computation Assertions that some event does eventually happen in the course of a computation Security/Trust The ability of a WS (composition) to provide proper authentication, authorization, confidentiality and data encryption The assurance that a WS (composition) will perform as expected despite possible environmental disruptions, human and operators errors, hostile attacks and design and implementation errors

WS Composition Characteristics III Quality of Service (QoS): Accuracy The error rate of a WS, measured as the number of errors generated by a WS in a certain time interval Availability The probability that a WS is available at any given time, measured as the percentage of time a WS is available over an extended time period Performance Measured as the success rate of WS requests: Maximum time needed to complete a request (response time) Number of completed requests over a period of time (throughput) Time needed by a WS to process a request (latency)

Comparison of Standardization Approaches Neither of these approaches offer any direct support for the verification of WS compositions at design time This is where formal methods come into play !

Formal Methods for WS Composition I Automata Well-known model underlying formal specifications I/O automata, timed automata, team automata, etc. Their formal basis allows for automatic tool support Exemplary approaches (see paper for references) Frameworks to analyze and verify properties of WS compositions of BPEL processes Translations from BPEL to Promela (finite automata) to use the SPIN model checker to verify LTL properties Translations from WS-CDL to timed automata to use the UPPAAL model checker to verify (timed) CTL properties

Formal Methods for WS Composition II Petri nets Well-known framework for modeling concurrent systems Their ease of conceptual modeling (graphical notation) has made Petri nets the model of choice in many applications Their formal basis allows for automatic tool support Exemplary approaches (see paper for references) Mapping of all BPEL control-flow constructs into labeled Petri nets (including the dead-path-elimination technique) Open-source tools BPEL2PNML and WofBPEL automatically transform BPEL processes in Petri nets and analyze them (including reachability analysis)

Formal Methods for WS Composition III Process Algebras Precise and well-studied set of formalisms CCS, π-calculus (which inspired BPEL to a certain extent), LOTOS, etc. Their formal basis allows automatic verification of behavioral properties Rich theory on bisimulation analysis for equivalence testing (to verify substitutivity and redundancy in WS compositions) Exemplary approaches (see paper for references) Specify and compose WSs in CCS to use Concurrency Workbench to validate correctness properties Translations from BPEL to LOTOS to use CADP model-checking toolbox to verify temporal properties

Comparing Formal Methods Paper provides a reference for WS composition designers and developers willing to use formal methods and tools

Conclusions Most standardization approaches to WS composition lack: Support to verify the (behavioral) correctness of WS compositions Support to perform quantitative analysis of QoS aspects Formal Methods and tools allow one to simulate and verify the behavior of one’s model at design time Thus enable the detection and correction of errors as early as possible and in any case before implementation ! The use of formal methods can increase the confidence in the correctness of one’s (WS composition) design