Abstract
RoCE is widely deployed in data centers for AI training, but network congestion causes GPU underutilization and slows training. Existing congestion control algorithms suffer from slow feedback loops and isolated handling of slowdown and recovery, leading to oscillations and poor bandwidth utilization. We propose CHOCOLATE, which integrates reverse explicit notification (REN) with lightweight flow prediction for congestion sensing, and unifies slowdown and recovery into a coordinated control system. Experimental results demonstrate improved RoCE performance in lossy networks.
Keywords
MSC
68M10, 68M12, 68W20
Cite this article
@article{cai-9380,
author = {Haotian Wu and Jiasheng Wang and Yuehao Xu and Song He and Jian Yu and Zhijun Ding},
title = {CHOCOLATE: Congestion Handling on Controllable Oblivious Lightweight Algorithm Through Explicit-Notification},
journal = {Computing and Informatics},
volume = {45},
number = {4},
pages = {995--1016},
year = {2026},
issn = {1335-9150},
doi = {10.31577/cai_2026_4_995}
}
TY - JOUR AU - Haotian Wu AU - Jiasheng Wang AU - Yuehao Xu AU - Song He AU - Jian Yu AU - Zhijun Ding TI - CHOCOLATE: Congestion Handling on Controllable Oblivious Lightweight Algorithm Through Explicit-Notification JO - Computing and Informatics VL - 45 IS - 4 SP - 995 EP - 1016 PY - 2026 SN - 1335-9150 DO - 10.31577/cai_2026_4_995 UR - https://doi.org/10.31577/cai_2026_4_995 KW - RDMA KW - RoCE KW - congestion control KW - data center network KW - switch ER -
Wu, H., Wang, J., Xu, Y., He, S., Yu, J., & Ding, Z. (2026). CHOCOLATE: Congestion Handling on Controllable Oblivious Lightweight Algorithm Through Explicit-Notification. Computing and Informatics, 45(4), 995–1016. https://doi.org/10.31577/cai_2026_4_995