Grid technology - Vision, architecture, and node capacity considerations

Grid computing has transitioned from its origin in research and development into the commercial arena. From a commercial perspective, Grid computing is composed of cluster systems that are linked together. Challenges faced by the Grid scheduler and the Grid resource managers are influenced by associated communication overhead. A Grid is a heterogeneous environment and hence some nodes in the Grid have a much higher capacity potential than other nodes. A scheduler may send a job to a saturated Grid node, which results in having to re-route the task. In a Grid environment, minimizing the job loss due to any latency issues can be achieved via specifying node specific thresholds. When reached, the node sends a message to the scheduler, instructing it to throttle back to fewer or zero jobs. In this study, an analytical model with processor sharing is introduced to quantify the capacity behavior of the Grid nodes, focusing on improving the reliability and aggregate performance behavior of Grid applications. The model is augmented by a Monte Carlo based probability estimation procedure that focuses on optimizing the communication behavior between a Grid node and the Grid scheduler, respectively.