摘要：The evaluation of the efficiency and robustness of scheduling in large scale distributed system are often achieved through the usage of Random Directed Acyclic Graphs (DAGs). However, the DAGs used for evaluating certain scheduling algorithms need to be generated taking into account a series of particular conditions for their structure. This is useful for generating DAGs whose structures are more similar to real application workflows and enhance a better evaluation of scheduling algorithms. Furthermore metrics, such as task granularity, communication to computation ratio, parallelism degree and constraints degree, that are used for evaluating DAGs are proposed along with an algorithm used in the same purpose.
