Efficient Matchmaking of Business Processes in Web Service Infrastructures

• Main Author: Mahleko, Bendick
• Format: Others
• Language: English; en
• Published: 2006
• Online Access: http://tuprints.ulb.tu-darmstadt.de/719/1/mahleko.pdf; Mahleko, Bendick <http://tuprints.ulb.tu-darmstadt.de/view/person/Mahleko=3ABendick=3A=3A.html> : Efficient Matchmaking of Business Processes in Web Service Infrastructures. [Online-Edition] Technische Universität, Darmstadt [Ph.D. Thesis], (2006)

The problem addressed in this dissertation can be stated as follows: given as a query, a business process description within a Web service infrastructure, efficiently find business processes from a large repository that complementarily support the input query. In particular, the input business process enforces that some parts of the business process be mandatory. For example, a buyer organization using RosettaNet PIPs and the corresponding data dictionary might want to find suppliers that use the same RosettaNet standard that fulfill his business process. In this case, for instance, a mandatory cancellation procedure of the processing would be executed after the order had been placed with the supplier. The current Web service standards around Universal Description, Discovery and Integration (UDDI) and related technologies do not address this problem as they offer only service discovery based on attribute/ value queries, that is limited to atomic service discovery. Research has shown that Business Process Execution Language for Web Services (BPEL4WS or BPEL for short), which is currently used to express business processes in Web service environments, does not have a solid formal model and thus lacks formal semantics for querying business process descriptions. This means that a formal model is required to express business processes and to enable their querying. Based on this formal model, appropriate indexing techniques are needed for efficient querying in large service repositories. A business process can be described as a set of message sequences each representing a potential execution sequence of the process. Based on this model, business processes are queried by finding matching business processes within the repository; the matching operation can be defined by computing a common set of message sequences covering the mandatory parts. The business process matchmaking semantics can be formalized by modelling business processes using an enrichment of finite state automata (FSAs) with propositional logical expressions. We call the enriched FSAs annotated finite state automata (aFSAs). aFSAs can express message sequences (potentially infinite) in a business process while logical expressions express mandatory as well as optional parts of the business process. A business process matchmaking is formally defined as the non-empty intersection of aFSAs, covering mandatory parts of the business process description. Computing the intersection of aFSAs is computationally expensive being more than quadratic on the number of states and transitions. Further, sequentially scanning large service repositories to find matching business processes will not scale even if individual query operations were simple. The traditional way to speed-up query performance operations in databases is to use indexes. However, indexes for intersection-based queries on aFSAs are also not supported by standard database indexing techniques such as B+-trees. An option is to construct an index based on aFSA message sequences and annotations, and using message sequence equivalence as well as the evaluation of annotations to realize intersection operations. However, the number of message sequences in an aFSA may be infinite due to cycles in the business process specification. In this dissertation a formal model for indexing and querying business processes is developed. The indexing approach is based on abstractions to transform aFSAs via their grammars into forms that can be indexed by available indexing mechanisms like B+-trees. The indexing approach is implemented and evaluated. Evaluation results show that the index outperforms sequential scanning by several orders of magnitude. An analysis of evaluation results also shows that the index scales well with increasing data set sizes and exhibits good computational properties for searching, being approximately linear on the number of aFSAs in the data collection.