LMLingkun Meng

Lingkun Meng

*This is an algorithmically generated author record based on publicly available information.

Lingkun Meng has a total of 56 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Taming Generation Quality and Latency for Text-to-Image Serving at the Edge

    by Sen Dong, Ke Cheng, Lingkun Meng, Sheng Zhang, Xiangyu Li, Dongxu Wang, Xiaohang Shi, Andong Zhu, Zhuzhong Qian

  • Rail: ReArranging Inter-GPU Links for GPU-Centric Clusters

    by Haixin Nan, Jun Xu, Peirui Cao, Zhaochen Zhang, Yizhi Wang, Zhehao Lin, Yuhang Li, Chengyuan Huang, Wentao Fan, Xiaohu Xu, et al.

    In modern GPU-centric clusters, large-scale AI training relies on two distinct communication domains: a high-bandwidth intra-node domain using proprietary interconnects (e.g., NVLink), and a scale-out inter-node network domain (e.g., RDMA). We observe that the widely-used ring algorithm, often create a significant load imbalance across these domains. This leads to the counter-intuitive scenario where the expensive, high-bandwidth intra-node domain becomes a performance bottleneck, while the inter-node network remains underutilized. This inefficiency is further exacerbated by the disparity in bandwidth provisioning: inter-node network bandwidth is generally more cost-effective and accessible, whereas intra-node bandwidth is often proprietary and more costly to scale. To address this fundamental imbalance, we propose Rail, aimed at resolving the intra-node bottleneck by strategically rearranging inter-GPU communication paths. This rebalancing ensures that traffic loads are appropriately matched with the distinct transmission capabilities of each domain, thereby maximizing overall communication performance. Rail incorporates a Load Distributing Strategy (LDS) that can accurately partition physical nodes into logical nodes based on the a transmission capabilities of both domains, shifting excess traffic from the overloaded intra-node domain to the underutilized network domain. Additionally, the Intra-Rail Strategy (IRS) leverages topological characteristics to ensure optimal communication paths through the network domain between logical nodes. Our evaluation demonstrates that Rail effectively mitigates congestion and achieves a 30.7% average increase in collective communication bus bandwidth compared to the widely-used NCCL solution.

  • UDMP: Unified Delay-Driven Multipath Protocol for AI Clusters

    by Chengyuan Huang, Zhengqi Cui, Jun Xu, Zhaochen Zhang, Li Wang, Peirui Cao, Zhongming Ji, Jilei Chen, Dongxu Wang, Shengju Zhang, et al.

    Distributed AI model training generates bursty, low-entropy elephant flows that challenge existing single-path transport protocols in multi-stage Clos networks, leading to congestion and inefficiency. Multipath transport emerges as a promising solution, leveraging multiple paths to balance traffic and enhance resilience. However, current multipath RDMA solutions suffer from scalability, congestion control, and load-balancing inefficiencies. This paper introduces Unified Delay-driven Multipath Protocol (UDMP), a novel approach that co-designs congestion control and load balancing using network delay as a unified signal. UDMP employs delay-gradient-based congestion control to precisely resolve unavoidable congestion. Moreover, UDMP leverages delay-assisted load balancing to shift traffic across paths with minimal latency adaptively, maintaining throughput when encountering avoidable congestion. A novel Token Pool design integrates these components, eliminating per-path state overhead while achieving fine-grained traffic distribution. Implementations on DPDK and NS3 demonstrate that UDMP achieves up to 2x higher throughput and reduces flow completion times by up to 30% compared to state-of-the-art methods like MPRDMA and QP-Scaling. These results highlight UDMP’s effectiveness in meeting the stringent performance requirements of modern distributed AI training workloads.


  • Open Access

    ERCC: Fine-grained RDMA Congestion Control via Kalman Filter-based Multi-bit ECN Feedback Reconstruction

    by Zhiguo Xu, Qingqiang Yi, Lingkun Meng, Shengju Zhang, Jilei Chen, Xinghua Zhao, Yunxiang Tao, Jun Xu, Qian Xue, Zhongming Ji, et al.


  • End-to-End Coordinated Spatio-Temporal Redundancy Elimination for Fast Video Analytics

    by Andong Zhu, Sheng Zhang, Lingkun Meng, Xiaohang Shi, Xiangyu Li, Dongxu Wang, Ke Cheng, Hesheng Sun, Sanglu Lu, Jie Wu, et al.

    Edge video analytics typically rely on conventional encoding standards to transmit device visual data for server-side inference. Unfortunately, general-purpose compression solutions retain unnecessary visual data that does not contribute to accuracy, resulting in significant latency throughout Video Analytics Pipeline (VAP). While previous approaches have made partial progress, they cannot systematically eliminate VAP redundancy due to uncoordinated subsystem-level optimization. Achieving complete redundancy elimination presents a major challenge, as a lack of spatio-temporal coordination risks offsetting latency gains with computational overhead (associated with redundancy elimination). Crucio overcomes these limitations with an end-to-edge framework that integrates temporally adaptive frame filtering and coordinated video compression. It leverages redesigned asymmetric autoencoders to synchronize inter-frame temporal compression with intra-frame spatial feature extraction. Additionally, Crucio employs a one-pass decoding mechanism for encoded critical frames and dynamically adjusts batching scales to minimize latency. Empirical results demonstrate Crucio’s superiority, outperforming existing solutions (e.g., DDS, Reducto, and STAC) by over a 31% reduction in end-to-end latency at 0.9 accuracy thresholds.


  • Mystique: User-Level Adaptation for Real-Time Video Analytics in Edge Networks via Meta-RL

    by Xiaohang Shi, Sheng Zhang, Meizhao Liu, Lingkun Meng, Liu Wei, Yingcheng Gu, Kai Liu, Huanyu Cheng, Yu Song, Lei Tang, et al.

    Deep neural network (DNN)-based real-time video analytics service, as a core module for numerous crucial applications such as augmented reality (AR), has garnered increasing research attention, where mobile edge computing (MEC) is often leveraged to mitigate its real-time processing burden on resource-constrained user devices. For Quality of Experience (QoE) optimization, latest works employ reinforcement learning (RL)-based methods to adaptively adjust configurations (e.g., resolution and frame rate), yet still presenting significant challenges. Firstly, we observe a substantial diversity in QoE patterns among users. Given that existing methods integrate a fixed QoE pattern in parameter training, it is intuitive to customize a policy network for each user. However, this necessitates significant training investment, failing to support on-the-fly deployment for new users. Secondly, given the dual dynamics from both the network and video content in edge video analytics system, existing methods often fall into the dilemma of fitting newly emerged and diverse system states with offline-trained fixed parameters. While it is promising to employ online learning algorithms, most of them struggle to catch up with the high dynamics. We hence propose Mystique. In real-time edge video analytics domain, it is the first meta-RL-based user-level configuration adaptation framework. Mystique establishes an initial model in offline meta training with model-agnostic meta-learning (MAML), enabling swift online adaptation to new users and system states through limited gradient updates from initial parameters. Comprehensive experiments illustrate that Mystique can improve QoE by 42% on average compared to prior works.