Trustworthy and Dynamic Mobile Task Scheduling in Data-Intensive Scientific Workflow Environments

摘要

There is an increasing demand for data-intensivernapplications in which scientists use scientific workflows tornintegrate together data management, analysis, simulation andrnvisualization services over often voluminous complex andrndistributed scientific data and services. One major limitationrnof current scientific workflow models is that each workflowrntask is stationary, requiring a dataset to be transferred from itsrnsource host to a target host where a stationary task residesrnbefore a computation can be performed on the dataset. Thisrnlimitation seriously impedes data-intensive applications sincernit can take an unbearable amount of time to transfer largernamount of datasets from their sources to the host where arnstationary task resides. In order to address this limitation, inrnthis paper, we apply the idea of mobile agents to distributedrnscientific workflows. In contrast to stationary tasks, mobilerntasks move from their home hosts towards datasets andrnperform computations on the dataset side. Since in dataintensivernapplications, it is often the case that the size of arnmobile task is much smaller than the size of a dataset, ourrnmobile-task approach can greatly reduce the networkrncommunication overhead. Since a mobile task might migraternacross various administrative domains and get executed atrnmultiple hosts, it is critically important to ensure the securityrnof a mobile-task-based workflow system. The lack of effectivernaccess control model creates a security hole in distributedrnscientific workflows. In this paper, we address this problemrnusing an itinerary-based access control model and a host visitrnscheduling algorithm that prevents arbitrary tasks fromrnaccessing and being executed on the current host.