In this paper, we study the design and analysis of online batching systems, which can be used to model many popular IT systems such as the Video-On-Demand system and the Pull-based Data Dissemination system. We analyze how the start-up delay affects the performance of these systems and investigate the tradeoff relationship between the number of servers and the length of the start-up delay. We also study a simple online algorithm for scheduling online batching systems. By deriving close lower and upper bounds, we prove that our algorithm is near optimal.