Consider a scenario where you need to train a linear regression model on a dataset with billions of records using Spark. Outline the steps you would take to ensure that the training process is both efficient and scalable, including data preprocessing, Spark configuration, and the use of Spark's MLlib functions.