arXiv subjects

Shuyi Pei

Publications and source records attributed to Shuyi Pei.

6 recordsLinked to original sources

arXiv2026

PLoRA: An NDP-Enhanced Pooled-Memory System for Cost-Efficient Multi-LoRA Serving

Multi-LoRA serving is how one base model becomes thousands of specialized variants, one adapter per user, task, or agent, and the deployments can hold 1000-plus adapters. Serving them is hard because the workload inverts what GPUs provide: terabytes of memory against only tens of TFLOPS, and because every published system stages its adapters from CPU DRAM over PCIe, where each access pays a kernel stop and a host-run copy and capacity ends at the motherboard's DIMM slots. Meanwhile, memory-semantic fabrics such as CXL and NVLink are converging on pooled memory that an accelerator addresses with its own loads and stores, and near-data processing (NDP) can place compute beside the pooled data. How to serve multi-LoRA workloads on such hardware remains unexplored. This paper introduces PLoRA, an NDP-enhanced pooled-memory system for cost-efficient multi-LoRA serving. PLoRA keeps adapters and KV cache in the pool and returns only reduced results over the link, through a read-compute interface the GPU drives with its own loads and stores. Above this architecture, a GPU memory management system picks among four LoRA and two attention execution strategies for each adapter and caches the most performance-critical bytes in GPU memory, guided by a link-parameterized cost model. On one H100 serving 1000 adapters, PLoRA attains the lowest decode latency on every model and workload we measure, averaging 6.6x below a real-machine S-LoRA at under 3.4% added device area. The link itself stops mattering: throughput saturates at 32 GB/s on short contexts, a quarter of CXL 3.1, and the verdict survives scale: per-GPU demand falls from 7B to a modeled 1.2T deployment once adapter traffic shards with the tensor parallelism. The design runs unchanged from CXL-class to NVLink-class fabrics, and surplus bandwidth buys pooled capacity rather than speed.

cs.AR↗

arXiv2026

Patterns behind Chaos: Forecasting Data Movement for Efficient Large-Scale MoE LLM Inference

Large-scale Mixture of Experts (MoE) Large Language Models (LLMs) have recently become the frontier open-weight models, achieving remarkable model capability similar to proprietary ones. But their random expert selection mechanism introduces significant data movement overhead that becomes the dominant bottleneck in multi-unit LLM serving systems. To understand the patterns underlying this data movement, we conduct comprehensive data-movement-centric profiling across four state-of-the-art large-scale MoE models released in 2025 (200B-1000B) using over 24,000 requests spanning diverse workloads. We perform systematic analysis from both temporal and spatial perspectives and distill six key insights to guide the design of diverse serving systems. We verify these insights on both future wafer-scale GPU architectures and existing GPU systems. On wafer-scale GPUs, lightweight architectural modifications guided by our insights yield a 6.6$\times$ average speedup across four 200B--1000B models. On existing GPU systems, our insights drive the design of a prefill-aware expert placement algorithm that achieves up to 1.25$\times$ speedup on MoE computation. Our work presents the first comprehensive data-centric analysis of large-scale MoE models together with a concrete design study applying the learned lessons. Our profiling traces are publicly available at \href{https://huggingface.co/datasets/core12345/MoE_expert_selection_trace}{\textcolor{blue}{https://huggingface.co/datasets/core12345/MoE\_expert\_selection\_trace}}.

cs.DC↗

arXiv2026

AMMA: A Multi-Chiplet Memory-Centric Architecture for Low-Latency 1M Context Attention Serving

All current LLM serving systems place the GPU at the center, from production-level attention-FFN disaggregation to NVIDIA's Rubin GPU-LPU heterogeneous platform. Even academic PIM/PNM proposals still treat the GPU as the central hub for cross-device communication. Yet the GPU's compute-rich architecture is fundamentally mismatched with the memory-bound nature of decode-phase attention, inflating serving latency while wasting power and die area on idle compute units. The problem is compounded as reasoning and agentic workloads push context lengths toward one million tokens, making attention latency the primary user-facing bottleneck. To address these inefficiencies, we present AMMA, a multi-chiplet, memory-centric architecture for low-latency long-context attention. AMMA replaces GPU compute dies with HBM-PNM cubes, roughly doubling the available memory bandwidth to better serve memory-bound attention workloads. To translate this bandwidth into proportional performance gains, we introduce (i) a logic-die microarchitecture that fully exploits per-cube internal bandwidth for decode attention under a minimal power and area budget, (ii) a two-level hybrid parallelism scheme, and (iii) a reordered collective flow that reduces intra-chip die-to-die communication overhead. We further conduct a design-space exploration over per-cube compute power and intra-chip D2D link bandwidth, providing actionable guidance for hardware designers. Evaluations show that AMMA achieves 15.5X lower attention latency and 6.9X lower energy consumption compared with the NVIDIA H100.

cs.AR↗

arXiv2025

Efficient Mixture-of-Agents Serving via Tree-Structured Routing, Adaptive Pruning, and Dependency-Aware Prefill-Decode Overlap

Mixture-of-Agents (MoA) inference can suffer from dense inter-agent communication and low hardware utilization, which jointly inflate serving latency. We present a serving design that targets these bottlenecks through an algorithm-system co-design. First, we replace dense agent interaction graphs with a hierarchical tree topology that induces structured sparsity in inter-agent communication. Second, we introduce a runtime adaptive mechanism that selectively terminates or skips downstream agent invocations using semantic agreement and confidence signals from intermediate outputs. Third, we pipeline agent execution by overlapping incremental prefilling with decoding across dependency-related agents, improving utilization and reducing inference latency. Across representative tasks, this approach substantially reduces end-to-end latency (up to 90%) while maintaining comparable accuracy (within $\pm$1%) relative to dense-connectivity MoA baselines, and can improve accuracy in certain settings.

cs.AI↗

arXiv2025

Preparation Meets Opportunity: Enhancing Data Preprocessing for ML Training With Seneca

Input data preprocessing is a common bottleneck when concurrently training multimedia machine learning (ML) models in modern systems. To alleviate these bottlenecks and reduce the training time for concurrent jobs, we present Seneca, a data loading system that optimizes cache partitioning and data sampling for the data storage and ingestion (DSI) pipeline. The design of Seneca contains two key techniques. First, Seneca uses a performance model for the data pipeline to optimally partition the cache for three different forms of data (encoded, decoded, and augmented). Second, Seneca opportunistically serves cached data over uncached ones during random batch sampling so that concurrent jobs benefit from each other. We implement Seneca by modifying PyTorch and demonstrate its effectiveness by comparing it against several state-of-the-art caching systems for DNN training. Seneca reduces the makespan by 45.23% compared to PyTorch and increases data processing throughput by up to 3.45x compared to the next best dataloader.

cs.OS↗

arXiv2025

Performance Characterizations and Usage Guidelines of Samsung CXL Memory Module Hybrid Prototype

The growing prevalence of data-intensive workloads, such as artificial intelligence (AI), machine learning (ML), high-performance computing (HPC), in-memory databases, and real-time analytics, has exposed limitations in conventional memory technologies like DRAM. While DRAM offers low latency and high throughput, it is constrained by high costs, scalability challenges, and volatility, making it less viable for capacity-bound and persistent applications in modern datacenters. Recently, Compute Express Link (CXL) has emerged as a promising alternative, enabling high-speed, cacheline-granular communication between CPUs and external devices. By leveraging CXL technology, NAND flash can now be used as memory expansion, offering three-fold benefits: byte-addressability, scalable capacity, and persistence at a low cost. Samsung's CXL Memory Module Hybrid (CMM-H) is the first product to deliver these benefits through a hardware-only solution, i.e., it does not incur any OS and IO overheads like conventional block devices. In particular, CMM-H integrates a DRAM cache with NAND flash in a single device to deliver near-DRAM latency. This paper presents the first publicly available study for comprehensive characterizations of an FPGA-based CMM-H prototype. Through this study, we address users' concerns about whether a wide variety of applications can successfully run on a memory device backed by NAND flash medium. Additionally, based on these characterizations, we provide key insights into how to best take advantage of the CMM-H device.

cs.AR↗