JWJinhui Wei

Jinhui Wei

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Jinhui Wei has a total of 26 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    StrataJ: Enhancing File System Journaling Scalability via Adaptive Concurrency Control and Full-Stack Optimizations

    by Ziyan Zhang, Yubo Liu, Yongfeng Wang, Yuanyin Zhang, Shaoyin Huang, Yuxiao Han, Jinhui Wei, Yizhuo Rao, Languang Gao, Zhiguang Chen, et al.

    The scalability of file system journaling faces significant challenges due to the increasing concurrency demands of modern multi-core CPUs and high throughput NVMe SSDs. However, our study reveals that the concurrency of journaling should be carefully adjusted to the workload, and the underlying I/O stack must be customized to better support concurrent journaling. To address these challenges, we propose StrataJ , a full-stack journaling system designed to enhance scalability through cross-layered optimizations spanning the journaling, block I/O, and driver layers. First, we design an adaptive journaling layer that dynamically adjusts the number of journaling streams based on the real-time analysis of workloads. Second, we propose a scalable I/O control mechanism in the block layer, which eliminates contention by parallelizing writeback throttling and physical region page (PRP) management. Third, we develop a decentralized I/O virtual address (IOVA) management scheme in the driver layer, which enables different regions of the address space to flush cache independently, thus in parallel. Our evaluations show that StrataJ significantly outperforms state-of-the-art systems across varied workloads, achieving up to 3.12×, 1.59× and 2.30× speedups on FxMark, Filebench, and RocksDB, respectively.

  • Open Access

    POLAR-PIC: A Holistic Framework for Matrixized PIC with Co-Designed Compute, Layout, and Communication

    by Yizhuo Rao, Xingjian Cui, Shangzhi Pang, Jiabin Xie, Guangnan Feng, Ziyan Zhang, Jinhui Wei, Languang Gao, Zhenyu Wang, Zhiguang Chen, et al.

  • Open Access

    Matrix‑PIC: Harnessing Matrix Outer-product for High‑Performance Particle‑in‑Cell Simulations

    by Yizhuo Rao, Xingjian Cui, Jiabin Xie, Shangzhi Pang, Guangnan Feng, Jinhui Wei, Zhiguang Chen, Yutong Lu

    Particle-in-Cell (PIC) simulations devote most cycles to particle-grid interactions, and their fine-grained atomic updates become a severe bottleneck on traditional many-core CPUs. The evolution of CPU architectures, particularly the integration of specialized Matrix Processing Units (MPUs) designed for efficient matrix outer-product operations, presents a paradigm shift and an opportunity to alleviate these bottlenecks. Capitalizing on this architectural advancement, this work focuses on adapting the critical current deposition step in PIC simulations to this new matrix-centric computational model. We introduce MatrixPIC, a novel framework representing, to our knowledge, the first holistic co-design of the deposition kernel, data layout, and an incremental sorting mechanism, all tailored specifically for the hybrid MPU-VPU SIMD execution model on modern CPUs. It provides a validated design paradigm for the scientific computing community to leverage the next generation of high-density hardware. The key technical innovations are: (i) the refactoring of the current deposition algorithm into block-matrix updates that map efficiently onto the MPU's native operational paradigm; (ii) a hybrid execution pipeline that synergistically orchestrates MPU kernels for high-density accumulation with VPU stages for data preparation and control flow; and (iii) an O(1)-amortized incremental particle sorter, which utilizes a gapped packed-memory array to establish and maintain the crucial data locality required for MPU efficiency. Evaluated on a next-generation high-performance computing platform, MatrixPIC demonstrates substantial performance improvements. In complex Laser-Wakefield Acceleration (LWFA) simulations, it yields up to a 2.63× speedup in total simulation time. Most significantly, for a higher-order, third-order deposition scheme, MatrixPIC accelerates the core deposition kernel by 8.7× over the original baseline and is 2.0× faster than our best hand-tuned VPU implementation. Furthermore, our CPU-based framework reaches 83.08% of theoretical peak performance, a hardware utilization rate nearly 2.8× higher than a highly-optimized CUDA kernel on a data center GPU. These results underscore the transformative potential of co-designing scientific algorithms for emerging, matrix-capable CPU architectures, paving the way for enhanced performance in large-scale computations.


  • Open Access

    Dynamic Latency-Throughput Balancing in Distributed Large Model Inference with Interleaved Parallelism

    by Jinhui Wei, Shenggan Cheng, Wei Zhu, Jiazhi Jiang, Dan Huang, Zhiguang Chen, Jiangsu Du, Yutong Lu

    Distributed large model inference is still in a dilemma of balancing cost and effect. Online scenarios require tensor parallelism to attain low latency, while the introduced intensive communications increase the cost. In contrast, pipeline parallelism enables high throughput with significantly reduced communication requirements, but it cannot improve each request’s effectiveness. Once a parallelism strategy is selected, the performance metrics become fixed, making it challenging to balance competing objectives. In this article, we present Liger+, a distributed large model inference system, that is, capable of achieving dynamic balance between latency and throughput on the multi-GPU architecture. The key idea lies in the novel interleaved parallelism, which interleaves the computation and communication across requests. Liger+ includes task-aware batch management and distributed runtime modules. The batch management module organizes batches based on features of discriminative and generative tasks and feeds the runtime module. The distributed runtime module strategically schedules computation and communication kernels across multiple requests onto multiple streams of multiple GPUs and enables the novel interleaved parallelism. First, it achieves precise control of kernel execution order efficiently by combining the CPU-GPU synchronization and the inter-stream synchronization. Second, it introduces the fine-grained resource mapping strategy and contention factor strategy to anticipate the penalty arising from resource contention. Third, it enables a higher degree of overlap by decomposing kernels into smaller, more manageable units at runtime. Extensive evaluations show that Liger+ can, in most cases, dynamically fit higher throughput demand and simultaneously achieve a better latency across models and devices, compared to fixed parallelism strategies. On a 4-device discriminative task, Liger+ reduces the P90 latency by 43.8% while maintaining the same throughput compared to the pipeline parallelism. Meanwhile, it improves the throughput by 1.53× with improved P90 latency compared to the tensor parallelism. For a 4-device generative task, Liger+ achieves an average 1.15× improvement in throughput and a 26.2% reduction in P90 latency compared to the tensor parallelism.

  • Open Access

    ASM-SpMM: Unleashing the Potential of Arm SME for Sparse Matrix Multiplication Acceleration

    by Jiazhi Jiang, Xijia Yao, Jiayu Chen, Jinhui Wei, Dan Huang, Yutong Lu

  • Open Access

    Liger: Interleaving Intra- and Inter-Operator Parallelism for Distributed Large Model Inference

    by Jiangsu Du, Jinhui Wei, Jiazhi Jiang, Shenggan Cheng, Dan Huang, Zhiguang Chen, Yutong Lu

    Distributed large model inference is still in a dilemma where balancing cost and effect. The online scenarios demand intraoperator parallelism to achieve low latency and intensive communications makes it costly. Conversely, the inter-operator parallelism can achieve high throughput with much fewer communications, but it fails to enhance the effectiveness. In this paper, we present Liger, a distributed large model inference runtime system that is of capability to achieve low latency at high throughput on the multi-GPU architecture. The key idea lies in the novel interleaved parallelism, which interleaves the computation and communication across requests. Liger enables this parallelism by carefully scheduling computation and communication kernels across requests onto multiple streams of multiple GPUs. It achieves precise control of kernel execution order efficiently by mixing use the CPU-GPU synchronization and the inter-stream synchronization. To prevent scheduling failures caused by resource contention, Liger introduces a contention factor strategy to anticipate the penalty of contention. It enables a higher degree of overlap by decomposing lengthy kernels into smaller, more manageable units at runtime. Extensive evaluations show that Liger, in most cases, outperforms existing parallelism approaches across models and devices, presenting the best latency and throughput results. In a 4-device case, Liger reduces the average latency by 36.0% while maintaining the same throughput compared to the inter-operator approach. Meanwhile, it improves the throughput by 1.34× with improved average latency compared to the intra-operator approach.