Continuous process improvement (CPI) may require a BPEL process to be split amongst different participants. In this paper, we enable splitting standard BPEL — without requiring any new middleware for the case of flat flows. The solution also supports splitting loops and scopes that have compensation and/or fault handlers. When splitting loops and scopes, we extend existing Web services standards and frameworks in a standard compliant manner in order to support the resulting split control (not data) between the fragments. Data dependencies, however, are handled directly using BPEL constructs placed in the fragments even for split loops and scopes. We present a solution that uses a BPEL process, partition information, and results of data-flow analysis to produce a BPEL process for each participant. The collective behavior of these participant processes recreates the control and data flow of the non-split process. Previous work presented process splitting using a variant of BPEL where data flow is modeled explicitly using "data links". We reuse the control flow aspect from that work as well as the control flow aspect from our work on splitting loops and scopes, focusing in this paper on maintaining the data dependencies in standard BPEL.
Early Results with Precision Abstraction: Using Data-flow Analysis to Improve the Scalability of Model Checking
