Key Technologies On Service Oriented Grid Workflow

Up to now grid workflow has been an increasingly attractive research branch in studies ongrid computing according to the fact that many relative projects have been initiated by theinstitutes or universities in Europe and U.S.A. Although some research goals have beenobtained there is still a long way to go since some critical problems featured in grid remain opennow, which include that there isn't an exact or efficient system architecture for grid workflow,and that its process model is not established soundly. The emergence of grid service broughtabout a novel research issue for grid study but many workflow projects couldn't follow such astep. Therefore, some key technologies for grid workflow have been studied in this paperincluding the architecture of grid service workflow (GSW), process modeling, servicescomposition and GSW's scheduling. The main contributions to grid workflow are listed asfollowing.At first a novel concept, grid service workflow, is defined for building a service orientedgrid workflow system, and then an architecture of GSW is proposed, which is composed ofworkflow engine, services composition module, services scheduling module and someinfrastructural services. In order to describe a GSW application, Grid Service WorkflowLanguage (GSWL) is also specified based on the improvements of GSFL. Also a specificationof a GSW instance is provided for verifying GSWL in the paper.Secondly, a process model of GSW is established based on Pi-Calculus, a branch ofprocess algebra, in which primary grid activities and relations among them are defined usingnames and processes of Pi-Calculus. Its goal is to verify the safety of a GSW instance duringdesign phase, so a safety verification algorithm is also issued as an analysis tool.Thirdly, addressing services composition in a GSW instance, protocol level service'sbehaviors are defined by its input, output and message orders at first, which comprise in theservice interaction patterns. Based on two elementary patterns all the grid services involved in asame GSW instance can form services context, a set of services in simple words. Thereforesome transition rules of Pi-Calculus are implemented in the context for verifying the correctnessof services composition, which is performed by executing PGSCV algorithm. For avoidingservice deadlock in a GSW instance, grid service adaptation is also investigated which results ina proposed service deadlock eliminating algorithm, PGSDE. All the experimental results provethat these two algorithms are efficient and available for grid services composition. Addressing the optimization problem in GSW, an optimized system is proposed thatbalances scheduling performance and cost and reduces execution time by parallel processing inmulti service domains. As a key to the system, an optimization algorithm, GSWGA, issuggested based on multi objective and parallel genetic algorithms, which produces a set ofPareto values as solutions of the GSW application at last.Also a prototype GSW system is designed and realized after the above research resultswere integrated and considered, whose goal is to perform the image processing in gridenvironment. Some codes of the system come from the open source projects including Globusand Triana. The experimental results prove that this system can work well for many GSWinstances.