Batched network coding (BNC) is a low-complexity variant of random linear network coding (RLNC) that encodes the file to be sent into batches, each consisting of a few coded packets. Although many works are focusing on the overhead reduction of RLNC, the overhead reduction for BNC is rarely studied. Different types of overheads in BNC interact and can potentially increase the overall overhead significantly. In this paper, we formulate a minimization problem for BNC on the number of symbols we need to send at the source node by tuning the number of packets divided from the file, which has the same meaning as tuning the packet size. It is hard to optimize the discrete zigzag-like objective efficiently. However, in practical scenarios, the encoder has to optimize the problem in real-time for different file size and redundancy requirements. We propose an efficient heuristic for this purpose and show its accuracy by numerical evaluations.