In the process of optimizing Spark performance for a machine learning project that involves graph data, which technique employs graph partitioning algorithms to enhance the distribution of graph nodes across cluster nodes?