To optimize performance in Apache Spark when joining a DataFrame df with another DataFrame lookupDf on a common key, which method should you use to leverage broadcast variables effectively?