Research On Key Issues Of Service Oriented Grid Workflow

Grid workflow is defined as the orchestration of a set of dependent atomic tasks processed at distributed resources in a well-defined order to accomplish a large and sophisticated goal. Due to the highly heterogeneous and dynamic features of grid, grid technologies are evolving towards service-oriented architecture. By encapsulating distributed heterogeneous hardware and software resources in standard interfaces, loosely coupled grid services such as Web services and WSRF services become dynamically discoverable and composable. Therefore, grid workflow can be composed of existing elementary services with sophisticated workflow logic. This paper presents the research on service oriented grid workflow from four aspects: workflow modeling, verification and optimization; workflow QoS; workflow job scheduling; and grid workflow system and applications.Frist, based on Petri nets, this paper presents a service composition oriented grid workflow model and its related six elementary workflow patterns: sequence, condition, iteration, concurrency, synchronization, and triggering. All these elementary workflow patterns can be nested and combined recursively. Therefore, any complex and sophisticated grid workflow models can be constructed by them. In addition, several grid workflow verification and optimization mechanisms are proposed to help common users to perform verification and optimization analysis, such as reachability and liveness checking, deadlock prevention.Second, this paper presents a virtual service based QoS oriented abstract grid workflow model. Virtual service is proposed to gather and delegate multiple concrete physical services providing equivalent functionality but diverse QoS capabilities. Therefore, grid services are organized at different granularities, which are Physical Service, Virtual Service, Atomic Service, and Composite Service. Rather than requiring that all participating services are static physical services, this paper proposes that virtual services can participate in workflow process to construct abstract grid workflow. During the workflow runtime, two phases of QoS-aware service selection are proposed for dynamic service bindings. Pre-matching phase is designed to choose competent physical services according to both of their static and dynamic resource capacities. This, in turn, generates the service pool within which the physical services may compete with each other. In QoS-based service selection phase, an extensible multi-criteria QoS model is presented. Synthesized QoS Selection Algorithm (SQSA) is proposed to quantitatively evaluate the quality of each candidate physical services and therefore the most qualified one can be selected.Third, based on list scheduling and group scheduling, this paper proposes a novel scheduling heuristic called DAGMap. DAGMap consists of two phases, namely Static Mapping and Dependable Execution. At the static mapping phase, firstly, based on dependency relationships and task upward priority, tasks are sorted into several groups. In each group, tasks are kept independent. Next, independent tasks are scheduled group by group in an ascending group order. Two features of independent tasks scheduling are: (1) Critical tasks are scheduled first; (2) Min-Min and Max-Min selective scheduling are used according to the length deviation of independent tasks. At the dependable execution phase, according to the static mapping results, tasks then are executed by dynamic grid resources. In order to provide fault tolerance, checkpoint server with cooperative checkpointing is proposed.Finally, this paper presents a service oriented grid workflow system called ServiceFlow, which support virtual services participating in abstract grid workflow. To reuse legacy software without reengineering its source codes, General Running Service (GRS) is proposed as a legacy program execution framework. through GRS, all legacy programs with its heterogeneous target systems can be encapsulated as standard Web service. Thus, legacy systems can be invoked remotely by SOAP message and integrated into workflow. In order to efficiently transfer the input/output data among atomic services, three data transfer approaches are provided: standard SOAP, SOAP attachment, and GridFTP. At last, to examine the service oriented grid workflow theory, two typical grid workflow applications are studied: (1) workflow based dependent software deployment; (2) workflow based image processing.