The development of workflow systems are returning from the scientificcomputing to the cloud services for normal users. Distributed workflowscheduling algorithms has become a hot research area of the distributedworkflow system design. Research on Scheduling large distributed workflowsbased on participants’ attributes is important and has potential applicationprospect.This thesis promote a process instance scheduling framework(ProcessInstance Scheduling Framework,PISF)consisting of interfaces. Each interfaceuses specific scheduling factors and makes decisions independently.Scheduling classes are introduced to allow business processes to specify theirscheduling algorithms. A load algorithm and a position algorithm areimplemented on PISF. The former balance the number of process instances bybroadcasting peers’ load in the P2P network formed by workflow engines. Thelatter moves instances to where participants locate to minimize thecommunication cost between workflow engines and their users.For supporting the distributed workflow scheduling, this thesis alsopromotes a process instance migration(Process Instance Migration,PIM)service; promotes a WebApp binding service to allow process instancesmigrated continue their support for original business systems; promotes aparticipant binding service to allow scheduling algorithms to use participants’attributes.At last, the experiment convinces the correctness of PISF, PIM and the twoservices mentioned. The result convinces the load algorithm can significantlyreduce the execution time of processes, and the effect is more obvious forprocesses not waiting for long responses from business systems. It alsoconvinces the position algorithm works. |