Yang, Lai, and Shih compile, in this paper, several of their conference publications on the design and implementation of a grid platform. This platform was developed as part of the TIGER project, and integrated resources distributed across five universities in Taichung City, Taiwan. The main focus of the paper is the development of a resource broker that can schedule jobs with dependencies that can be expressed as a directed acyclic graph (DAG) on a set of available resources. In scheduling the jobs, the broker also considers the expected network and central processing unit (CPU) requirements.
The authors describe the integration work for building the platform that involves several well-known software toolkits and products. They also propose a history-based execution time estimation model to predict the execution time of parallel applications, and provide experimental results that show its utility. Experimental results of using the broker are included.
I found the use of the word “workflow” in the title (and in the text) a bit misleading. A workflow, as it relates to scheduling, involves a work process that is executed repeatedly. One could, therefore, expect that a workflow-based broker could map a DAG or part of a DAG to a set of available machines, make sure that the appropriate output and input nodes are connected, and monitor the execution to maintain quality of service. However, the paper only describes scheduling jobs with DAG-like dependencies, using a topological sort-based order. Overall, the paper is easy to read and follow.