arXiv subjects

Guyue Liu

Publications and source records attributed to Guyue Liu.

13 recordsLinked to original sources

arXiv2026

HetCCL: Enabling Collective Communication For Mixed-Vendor Heterogeneous Clusters

Training Large Language Models (LLMs) on heterogeneous clusters presents significant challenges for collective communication, as hardware from multiple vendors introduces diverse network and computational characteristics. Existing collective communication frameworks (e.g., NCCL, RCCL) designed for homogeneous environments fail to address mixed-hardware setups, while communication libraries with heterogeneous support (e.g., Gloo, OpenMPI) incur heavy overhead in the data path. This paper presents HetCCL, a framework that enables heterogeneous collective communication by efficient P2P transport across heterogeneous devices (e.g., GPUs), eliminating the host-device memory copy overhead while offloading the control to the CPUs. For combining collectives (e.g., AllReduce, ReduceScatter), HetCCL introduces a border-communicator mechanism that achieves vendor independence by using the intrinsic reduction in the combining collectives in vendor collective communication libraries. With efficient heterogeneous P2P transport and portable reduction mechanism, HetCCL proposes a hierarchical topology abstraction for heterogeneous clusters, dissecting collective communication into cluster-level primitives that guarantee optimal cross-cluster data transfer volume and optimal bandwidth utilization. We implement HetCCL with 4 different vendor support and evaluate it in 4 heterogeneous settings with benchmarks and end-to-end LLM tasks. Our evaluation shows that HetCCL achieves 17-19x higher bandwidth than Gloo in heterogeneous communications, and speeds up end-to-end training by up to 16.9% in the per-step-time.

cs.NI↗

arXiv2026

RelayCaching: Accelerating LLM Collaboration via Decoding KV Cache Reuse

The increasing complexity of AI tasks has shifted the paradigm from monolithic models toward multi-agent large language model (LLM) systems. However, these collaborative architectures introduce a critical bottleneck: redundant prefill computation for shared content generated by previous agents, which significantly increases KV cache memory usage and time-to-first-token (TTFT). While various KV cache methods have been proposed to mitigate prefill redundancy, they either fail to maintain accuracy on agent-generated outputs or exhibit low reuse rates due to rigid constraints. We present RelayCaching, a training-free inference method that directly reuses decoding phase KV caches from previous agents in subsequent prefill phases. Our key insight is that KV caches for identical content are highly consistent across phases, while prefix-induced deviations are sparse and localized within a limited range of layers and token positions. By selectively recomputing KV caches at these positions, RelayCaching preserves model accuracy with minimal overhead, yielding a superior accuracy-efficiency trade-off over existing methods. Experiments on diverse collaborative LLM tasks spanning mathematical reasoning, general knowledge, and code generation demonstrate that RelayCaching achieves over 80% KV cache reuse, reduces TTFT by up to $4.7\times$ compared to the standard pipeline, all with negligible accuracy degradation.

cs.LG↗

arXiv2026

XLB: A High Performance Layer-7 Load Balancer for Microservices using eBPF-based In-kernel Interposition

L7 load balancers are a fundamental building block in microservices as they enable fine-grained traffic distribution. Compared to monolithic applications, microservices demand higher performance and stricter isolation from load balancers. This is due to the increased number of instances, longer service chains, and the necessity for co-location with services on the same host. Traditional sidecar-based load balancers are ill-equipped to meet these demands, often resulting in significant performance degradation. In this work, we present XLB, a novel architecture that reshapes L7 load balancers as in-kernel interposition operating on the socket layer. We leverage eBPF to implement the core load balancing logic in the kernel, and address the connection management and state maintenance challenges through novel socket layer redirection and nested eBPF maps designs. XLB eliminates the extra overhead of scheduling, communication, and data movement, resulting in a more lightweight, scalable, and efficient L7 load balancer architecture. Compared to the widely used microservices load balancers (Istio and Cilium), over 50 microservice instances, XLB achieves up to 1.5x higher throughput and 60% lower end-to-end latency.

cs.NI↗

arXiv2025

Analyzing Communication Predictability in LLM Training

Effective communication is essential in distributed training, with predictability being one of its most significant characteristics. However, existing studies primarily focus on exploiting predictability through online profiling for runtime optimization, without a systematic understanding of it. In this work, we aim to systematically formulate communication predictability in distributed training, particularly in Large Language Models (LLMs) that utilize hybrid parallelism. Our analysis focuses on both traffic patterns and communication overhead. Specifically, we investigate predictable traffic patterns in typical LLMs and evaluate how various factors influence GPU utilization and effective bandwidth (two critical variables affecting communication overhead). Furthermore, we develop an analytical formulation to estimate communication overhead in LLM training, which is validated with high accuracy against empirical data. Leveraging this formulation, we propose a configuration tuning tool, ConfigTuner, to optimize training performance. Compared to Megatron-LM, the training configurations optimized by ConfigTuner demonstrate up to a 1.36$\times$ increase in throughput. Compared to Alpa, ConfigTuner generates the same configuration suggestion while significantly reducing the search complexity.

cs.NI↗

arXiv2025

xLLM Technical Report

We introduce xLLM, an intelligent and efficient Large Language Model (LLM) inference framework designed for high-performance, large-scale enterprise-grade serving, with deep optimizations for diverse AI accelerators. To address these challenges, xLLM builds a novel decoupled service-engine architecture. At the service layer, xLLM-Service features an intelligent scheduling module that efficiently processes multimodal requests and co-locates online and offline tasks through unified elastic scheduling to maximize cluster utilization. This module also relies on a workload-adaptive dynamic Prefill-Decode (PD) disaggregation policy and a novel Encode-Prefill-Decode (EPD) disaggregation policy designed for multimodal inputs. Furthermore, it incorporates a distributed architecture to provide global KV Cache management and robust fault-tolerant capabilities for high availability. At the engine layer, xLLM-Engine co-optimizes system and algorithm designs to fully saturate computing resources. This is achieved through comprehensive multi-layer execution pipeline optimizations, an adaptive graph mode and an xTensor memory management. xLLM-Engine also further integrates algorithmic enhancements such as optimized speculative decoding and dynamic EPLB, collectively serving to substantially boost throughput and inference efficiency. Extensive evaluations demonstrate that xLLM delivers significantly superior performance and resource efficiency. Under identical TPOT constraints, xLLM achieves throughput up to 1.7x that of MindIE and 2.2x that of vLLM-Ascend with Qwen-series models, while maintaining an average throughput of 1.7x that of MindIE with Deepseek-series models. xLLM framework is publicly available at https://github.com/jd-opensource/xllm and https://github.com/jd-opensource/xllm-service.

cs.DC↗

arXiv2025

Zoomer: Adaptive Image Focus Optimization for Black-box MLLM

Multimodal large language models (MLLMs) such as GPT-4o, Gemini Pro, and Claude 3.5 have enabled unified reasoning over text and visual inputs, yet they often hallucinate in real world scenarios especially when small objects or fine spatial context are involved. We pinpoint two core causes of this failure: the absence of region-adaptive attention and inflexible token budgets that force uniform downsampling, leading to critical information loss. To overcome these limitations, we introduce Zoomer, a visual prompting framework that delivers token-efficient, detail-preserving image representations for black-box MLLMs. Zoomer integrates (1) a prompt-aware emphasis module to highlight semantically relevant regions, (2) a spatial-preserving orchestration schema to maintain object relationships, and (3) a budget-aware strategy to adaptively allocate tokens between global context and local details. Extensive experiments on nine benchmarks and three commercial MLLMs demonstrate that Zoomer boosts accuracy by up to 27% while cutting image token usage by up to 67%. Our approach establishes a principled methodology for robust, resource-aware multimodal understanding in settings where model internals are inaccessible.

cs.CV↗

arXiv2025

NetCloak: Dynamic Topology Expansion for Secure and Scalable Configuration Sharing

As modern networks continue to grow in both scale and complexity, sharing real-world device configurations poses significant privacy risks, especially when adversaries can infer organizational size or resource distribution from topology data. We present NetCloak, a configuration anonymization framework that adaptively injects synthetic routers and hosts into the network graph to obfuscate true scale, while preserving end-to-end forwarding behavior. NetCloak core techniques include: (1) a graph-embedding expansion algorithm that integrates the original topology into a larger reference graph, ensuring added nodes blend seamlessly with real ones; (2) a k-degree mapping anonymity scheme that selectively adds minimal links to guarantee each original node degree is indistinguishable among at least k peers; (3) a mimicry-driven configuration generator that derives command templates from existing devices, preserving command ordering, naming conventions, and routing policies; and (4) a layered repair process combining SMT-based intra-AS route synthesis with iterative inter-AS filter insertion to restore protocol-correct routing under OSPF and BGP. Extensive experiments on real and emulated campus and data-center topologies demonstrate that NetCloak effectively conceals network size, improving topological rationality by over 70% and configuration fidelity by nearly 30% compared to baseline methods, while reducing route-repair overhead by more than 50% under randomized link costs. NetCloak thus enables safe, privacy-preserving configuration sharing at scale.

cs.NI↗

arXiv2025

InfiniteHBD: Building Datacenter-Scale High-Bandwidth Domain for LLM with Optical Circuit Switching Transceivers

Scaling Large Language Model (LLM) training relies on multi-dimensional parallelism, where High-Bandwidth Domains (HBDs) are critical for communication-intensive parallelism like Tensor Parallelism. However, existing HBD architectures face fundamental limitations in scalability, cost, and fault resiliency: switch-centric HBDs (e.g., NVL-72) incur prohibitive scaling costs, while GPU-centric HBDs (e.g., TPUv3/Dojo) suffer from severe fault propagation. Switch-GPU hybrid HBDs (e.g., TPUv4) take a middle-ground approach, but the fault explosion radius remains large. We propose InfiniteHBD, a transceiver-centric HBD architecture that integrates connectivity and dynamic switching at the transceiver level by embedding Optical Circuit Switching (OCS) within each transceiver. It enables reconfigurable point-to-multipoint communication and scalable variable-size ring topologies. InfiniteHBD achieves datacenter-scale scalability without cost explosion, fault isolation at the node level, and full bandwidth utilization for healthy GPUs. Key innovations include a Silicon Photonic-based OCS transceiver (OCSTrx), a reconfigurable k-hop ring topology, and an HBD-DCN orchestration algorithm. The evaluation demonstrates that InfiniteHBD reduces cost to 31% of NVL-72, achieves a near-zero GPU waste ratio (over 10x lower than NVL-72 and TPUv4), maintains near-zero cross-ToR traffic under 7% node fault ratio, and improves Model FLOPs Utilization by 3.37x compared to NVIDIA DGX (8 GPUs/node).

cs.NI↗

arXiv2025

MixNet: A Runtime Reconfigurable Optical-Electrical Fabric for Distributed Mixture-of-Experts Training

Mixture-of-Expert (MoE) models outperform conventional models by selectively activating different subnets, named experts, on a per-token basis. This gated computation generates dynamic communications that cannot be determined beforehand, challenging the existing GPU interconnects that remain static during the distributed training process. In this paper, we advocate for a first-of-its-kind system, called MixNet, that unlocks topology reconfiguration during distributed MoE training. Towards this vision, we first perform a production measurement study and show that the MoE dynamic communication pattern has strong locality, alleviating the requirement of global reconfiguration. Based on this, we design and implement a regionally reconfigurable high-bandwidth domain on top of existing electrical interconnects using optical circuit switching (OCS), achieving scalability while maintaining rapid adaptability. We have built a fully functional MixNet prototype with commodity hardware and a customized collective communication runtime that trains state-of-the-art MoE models with in-training topology reconfiguration across 32 A100 GPUs. Large-scale packet-level simulations show that MixNet delivers comparable performance as the non-blocking fat-tree fabric while boosting the training cost efficiency (e.g., performance per dollar) of four representative MoE models by 1.2x-1.5x and 1.9x-2.3x at 100 Gbps and 400 Gbps link bandwidths, respectively.

cs.NI↗

arXiv2024

Large Language Model-Brained GUI Agents: A Survey

GUIs have long been central to human-computer interaction, providing an intuitive and visually-driven way to access and interact with digital systems. The advent of LLMs, particularly multimodal models, has ushered in a new era of GUI automation. They have demonstrated exceptional capabilities in natural language understanding, code generation, and visual processing. This has paved the way for a new generation of LLM-brained GUI agents capable of interpreting complex GUI elements and autonomously executing actions based on natural language instructions. These agents represent a paradigm shift, enabling users to perform intricate, multi-step tasks through simple conversational commands. Their applications span across web navigation, mobile app interactions, and desktop automation, offering a transformative user experience that revolutionizes how individuals interact with software. This emerging field is rapidly advancing, with significant progress in both research and industry. To provide a structured understanding of this trend, this paper presents a comprehensive survey of LLM-brained GUI agents, exploring their historical evolution, core components, and advanced techniques. We address research questions such as existing GUI agent frameworks, the collection and utilization of data for training specialized GUI agents, the development of large action models tailored for GUI tasks, and the evaluation metrics and benchmarks necessary to assess their effectiveness. Additionally, we examine emerging applications powered by these agents. Through a detailed analysis, this survey identifies key research gaps and outlines a roadmap for future advancements in the field. By consolidating foundational knowledge and state-of-the-art developments, this work aims to guide both researchers and practitioners in overcoming challenges and unlocking the full potential of LLM-brained GUI agents.

cs.AI↗

arXiv2023

ClickINC: In-network Computing as a Service in Heterogeneous Programmable Data-center Networks

In-Network Computing (INC) has found many applications for performance boosts or cost reduction. However, given heterogeneous devices, diverse applications, and multi-path network typologies, it is cumbersome and error-prone for application developers to effectively utilize the available network resources and gain predictable benefits without impeding normal network functions. Previous work is oriented to network operators more than application developers. We develop ClickINC to streamline the INC programming and deployment using a unified and automated workflow. ClickINC provides INC developers a modular programming abstractions, without concerning to the states of the devices and the network topology. We describe the ClickINC framework, model, language, workflow, and corresponding algorithms. Experiments on both an emulator and a prototype system demonstrate its feasibility and benefits.

cs.NI↗

arXiv2019

Efficient, Dynamic Multi-tenant Edge Computation in EdgeOS

In the future, computing will be immersed in the world around us -- from augmented reality to autonomous vehicles to the Internet of Things. Many of these smart devices will offer services that respond in real time to their physical surroundings, requiring complex processing with strict performance guarantees. Edge clouds promise a pervasive computational infrastructure a short network hop away from end devices, but today's operating systems are a poor fit to meet the goals of scalable isolation, dense multi-tenancy, and predictable performance required by these emerging applications. In this paper we present EdgeOS, a micro-kernel based operating system that meets these goals by blending recent advances in real-time systems and network function virtualization. EdgeOS introduces a Featherweight Process model that offers lightweight isolation and supports extreme scalability even under high churn. Our architecture provides efficient communication mechanisms, and low-overhead per-client isolation. To achieve high performance networking, EdgeOS employs kernel bypass paired with the isolation properties of Featherweight Processes. We have evaluated our EdgeOS prototype for running high scale network middleboxes using the Click software router and endpoint applications using memcached. EdgeOS reduces startup latency by 170X compared to Linux processes and over five orders of magnitude compared to containers, while providing three orders of magnitude latency improvement when running 300 to 1000 edge-cloud memcached instances on one server.

cs.OS↗

arXiv2016

SDNFV: Flexible and Dynamic Software Defined Control of an Application- and Flow-Aware Data Plane

Software Defined Networking (SDN) promises greater flexibility for directing packet flows, and Network Function Virtualization promises to enable dynamic management of software-based network functions. However, the current divide between an intelligent control plane and an overly simple, stateless data plane results in the inability to exploit the flexibility of a software based network. In this paper we propose SDNFV, a framework that expands the capabilities of network processing-and-forwarding elements to flexibly manage packet flows, while retaining both a high performance data plane and an easily managed control plane. SDNFV proposes a hierarchical control framework where decisions are made across the SDN controller, a host-level manager, and individual VMs to best exploit state available at each level. This increases the network's flexibility compared to existing SDNs where controllers often make decisions solely based on the first packet header of a flow. SDNFV intelligently places network services across hosts and connects them in sequential and parallel chains, giving both the SDN controller and individual network functions the ability to enhance and update flow rules to adapt to changing conditions. Our prototype demonstrates how to efficiently and flexibly reroute flows based on data plane state such as packet payloads and traffic characteristics.

cs.NI↗