You are analyzing the performance of a Spark streaming job using the Spark UI and notice that the event timeline shows a high number of late batches. What steps would you take to address this issue?