Preprint
Sep 2026
Accurate Simulation of Distributed Training Jobs with Network Contention Modeling
MoSim is introduced, a GPU-cluster simulator that models DT job execution under dynamic network contention, which reduces simulation error for average JCT by up to 3.28%, and reduces input construction overhead by 44.6% by avoiding real-GPU profiling.
Yeonho Yoo, Hyunho Lee, Hyun-Eem Choi et al.
· 0 citations