Exploring the Volatility of Large-Scale Shared Distributed Computing Resources

Scientific applications often require colossal amount of computing resources for running user’s tasks. Grid computing has been proved to be powerful research testbed for accessing massive amount of computing resources at almost zero cost across various autonomous administrative institutes. It can seamlessly integrate hundreds of thousands of geographically distributed heterogeneous computing resources from multiple domains organized into virtual organization (VO). Unfortunately, existing Grid Information Service (GIS) suffers from providing exact dynamic resource information due to its scale and autonomous resource management policies. In this paper, we present a comprehensive volatility study of shared computing resources by VO in terms of characterizing resource performance related features of each computing element (CE) in computational Grids such as number of available CPU cores, average response time. We also performed experiments based on a large number of micro-benchmark tasks on real Grid environment to analyze implication of resources fluctuation. Evaluation results reveal that resource volatility studies tremendously help to decrease user response time and job completion rate.