Agile Data Streaming for Grid Applications

The grid is proposed to be the new computing infrastructure for cross-domain resource sharing and collaboration in virtual organizations. While most current work in grid computing community focus on adaptability of grid resource management, agility is proposed in this work as another essential requirement, aiming to provide mass customization and personalization on grid computing service provision. This work is focused on data streaming aspects of grid agility. Since volumes of data streams are usually extremely high but available bandwidth and storage are often very limit in a grid environment, an enabling environment for grid data streaming applications is implemented with supports of on-demand data transfers and just-in-time data cleanups in this work. The implementation includes performance sensors, predictors and a scheduler, with real-time measurements, prediction and scheduling capabilities. The GridFTP is utilized and data transfer schedules are performed using the on-the-fly adjustable GridFTP parallelism. Experimental results show that the data streaming environment is agile enough to meet dynamically changing application data processing requirements and scale well regarding to storage and bandwidth usage.