StageFrontier: Synchronization-Aware Stage Accounting for Distributed ML Training
For operators of distributed training, StageFrontier offers a practical, low-overhead diagnostic tool to pinpoint slowdowns without the cost of heavy profilers.
StageFrontier introduces an always-on, lightweight signal that provides exact, additive accounting of exposed time in distributed ML training, identifying the stage and rank where group-visible delay first appears with under 0.2% throughput overhead on 128 ranks, and recovers the same top-stage routing as full profilers from a 0.11 MB summary instead of a 15.81 GB trace.
When a distributed training job slows down, the hard part is knowing where to look. Synchronization hides the cause: a stall on one rank shows up as a wait on the others, so a data delay on a single rank can surface as backward time across the group. The cheap dashboards that run all the time -- per-stage averages and maxima -- misread this, double-counting the same exposed delay or burying the slow rank in an average, while full profilers see it clearly but are far too heavy to leave on. StageFrontier is an always-on signal that closes this gap. Each rank reports only a short ordered vector of coarse stage durations -- data, forward, backward, and so on -- timed with CPU wall-clock, with no synchronized clocks and no kernel tracing. At each stage boundary, StageFrontier takes the cumulative time of whichever rank is furthest along; the increments of this frontier form an exact, additive accounting of the step's exposed time and point to the stage and rank where group-visible delay first appears, telling an operator where to aim a heavy profiler, not which fix to make. The accounting is exact, but the coarse signal alone cannot tell whether a leading stage truly caused the slowdown or merely ran alongside it; StageFrontier labels the windows where that distinction needs more evidence instead of guessing. A PyTorch implementation adds under 0.2% throughput overhead through 128 ranks on Gloo and NCCL, places injected faults among its top two suspects on all 50 rows of a hidden-rank DDP test, and recovers the same top-stage routing as PyTorch Profiler, HTA, and Nsight Systems once their traces are reduced to the same coarse stages -- from a 0.11 MB summary instead of a 15.81 GB trace.