High resource overhead and latency during data redistribution across compute nodes. Reducing these bottlenecks prevents network congestion and minimizes idle compute time.