MLTCP: Congestion Control for DNN Training
This addresses network inefficiencies for DNN training in shared clusters, offering a practical improvement with incremental modifications to existing protocols.
The paper tackles the problem of network congestion in shared GPU clusters during DNN training by introducing MLTCP, a technique that modifies congestion control algorithms to interleave communication phases, resulting in up to 2x faster average and 4x faster 99th percentile training iteration times.
We present MLTCP, a technique to augment today's congestion control algorithms to accelerate DNN training jobs in shared GPU clusters. MLTCP enables the communication phases of jobs that compete for network bandwidth to interleave with each other, thereby utilizing the network efficiently. At the heart of MLTCP lies a very simple principle based on a key conceptual insight: DNN training flows should scale their congestion window size based on the number of bytes sent at each training iteration. We show that integrating this principle into today's congestion control protocols is straightforward: by adding 30-60 lines of code to Reno, CUBIC, or DCQCN, MLTCP stabilizes flows of different jobs into an interleaved state within a few training iterations, regardless of the number of competing flows or the start time of each flow. Our experiments with popular DNN training jobs demonstrate that enabling MLTCP accelerates the average and 99th percentile training iteration time by up to 2x and 4x, respectively.