A data scientist is developing a recommendation system in Databricks using collaborative filtering. They have a user-item interaction DataFrame and want to train a model using the Alternating Least Squares (ALS) algorithm. What consideration is crucial for optimizing ALS performance?