In this paper, we show our current work on the relationships between local behaviors of agents and global performance of multi-agent systems. We conduct our experiments on RoboNBA, which is a multi-agent system testbed. Local behaviors and global performance in RoboNBA are introduced. In addition, we address the problem of how to quantitatively measure the global performance in RoboNBA. Through experiments and analysis, we try to examine how agents' local behaviors can lead to interesting global performance of a match (e.g., optimized match results) in three problems: (1) cooperation between agents; (2) rational decision making; (3) coordination among agents.