Many cloud applications in modern data centers are very demanding on latency, thus researchers have paid much attention to building a data center network with ultra-low latency, even deterministic low latency. Motivated by this research trend, this article focuses on effective latency reduction designs, which mainly aim at reducing the queuing delay in switches. We first summarize a development roadmap to give an overview of the three developing stages of existing schemes, and point out the essential difference between these stages is the amount of traffic information maintained by the control nodes. After briefly reviewing the features and deployment complexity of recent advances, we map them to three stages, introducing their design principles and identifying the problems they target at. Finally, we present the challenges and opportunities for future work.