Study of Grid Locality and Its Optimization

Data grid is playing an important role in scientific research work. Essentially, data grid is an infrastructure that manages large scale data sets and provides computational resources across widely distributed communities. It has been a research hotspot to improve the performance of data grid platform used to handle and manage large distributed data files. Data file replacements due to limited local storage and grid job assignments are key elements to the efficient data grid platform. Presented here is the concept of grid locality, which involves in job locality and file locality. And the impact on the performance of data grid produced by the grid locality is analyzed. Further, the performance improvement of data grid platform is studied in the perspective of grid locality. Considering the enhancement of grid locality, a composite policy focusing on file replacement and job assignment is put forward, which is jump-DRP (jump-diffusion replacement policy), as well as JARIP (job assignment referencing to insect pheromone). Jump-DRP is based on jump-diffusion features and JARIP is based on insect pheromone characteristics. Application simulation is carried out at the file access patterns of sequential, unitary random walk and Gaussian random walk, whereas job submission is in accordance with Gaussian distribution. Experiment results show the integration of jump-DRP and JARIP is robust for both various applications and diverse jobs on data grid platform.