arXiv subjects

Haipeng Yao

Publications and source records attributed to Haipeng Yao.

6 recordsLinked to original sources

arXiv2026

LCMP: Distributed Long-Haul Cost-Aware Multi-Path Routing for Inter-Datacenter RDMA Networks

RDMA-empowered cloud services are gradually deployed across datacenters (DCs) with multiple paths, which exhibit new properties of path asymmetry, delayed congestion signals, and simultaneous flow routing collisions, and further fail existing routing methods. We present LCMP, a distributed long-haul cost-aware multi-path routing framework that aims to place RDMA flows on multiple inter-DC paths, achieving low-cost, low-latency, and congestion-responsive transmission. LCMP combines a control-plane path-quality score with compact on-switch congestion signals, where the former unifies quality assessment for asymmetric paths and the latter enables responsive reaction to path congestion. LCMP further resolves the simultaneous flow decision collision problem by filtering high-cost candidates, and performing a diversity-preserving hash inside the reduced set. On an 8-DC testbed, LCMP reduces median and tail FCT slowdown by up to 76% and 64%, respectively compared to state-of-the-art (SOTA) DCN routing strategies. And large-scale NS-3 simulations under the 2000 km inter-DC scenario confirm similar improvements.

cs.NI↗

arXiv2025

SplitFrozen: Split Learning with Device-side Model Frozen for Fine-Tuning LLM on Heterogeneous Resource-Constrained Devices

Fine-tuning large language models (LLMs) on private, on-device data can empower tailored personalized AI agents. However, fine-tuning LLMs on resource-constrained edge devices faces significant challenges, including excessive computation overhead, device heterogeneity, and data imbalance. This paper proposes SplitFrozen, a split learning framework that enables efficient LLM fine-tuning by strategically freezing device-side model layers while centralizing parameter-efficient fine-tuning on the server. Our framework partitions LLMs into device-side frozen layers and server-side fine-tuning layers, where heterogeneous resource-constrained devices execute only forward propagation. To minimize server-side training costs, we integrate Low-Rank Adaptation (LoRA) into the server-side layers. A pipeline parallelism strategy further optimizes training efficiency by decoupling device-server computations and leveraging decomposed backward propagation. Experiments on GPT-2 with the MRPC, MNLI-matched, and SST-2 datasets demonstrate that SplitFrozen outperforms FedLoRA and SplitLoRA by 69.4\% model accuracy under extremely imbalanced data, while reducing up to 86.8\% device-side computations and 50.2\% total training time. Experiments also validate the scalability of SplitFrozen on content generation task using Llama-3.2 model on GSM8K dataset.

cs.LG↗

arXiv2022

A multi-domain virtual network embedding algorithm with delay prediction

Virtual network embedding (VNE) is an crucial part of network virtualization (NV), which aims to map the virtual networks (VNs) to a shared substrate network (SN). With the emergence of various delay-sensitive applications, how to improve the delay performance of the system has become a hot topic in academic circles. Based on extensive research, we proposed a multi-domain virtual network embedding algorithm based on delay prediction (DP-VNE). Firstly, the candidate physical nodes are selected by estimating the delay of virtual requests, then particle swarm optimization (PSO) algorithm is used to optimize the mapping process, so as to reduce the delay of the system. The simulation results show that compared with the other three advanced algorithms, the proposed algorithm can significantly reduce the system delay while keeping other indicators unaffected.

cs.NI↗

arXiv2018

LEOS-assisted Inter-GEOS Communication via Distributed-storage Coding

We consider a space communication network consisting of Geosynchronous Earth Orbit satellites (GEOSs) and Low Earth Orbit satellites (LEOSs). In case of no direct communication link between two GEOSs, the data exchange between them is through relay by the LEOSs. In particular, the source GEOS sends coded data to multiple LEOSs based on the distributed storage framework. The destination GEOS then retrieves certain amount of data from each LEOS for data reconstruction. For the GEOS-LEOS downlink, a regenerating-code-based transmission scheme is optimized to guarantee data reconstructability, where the transmission power allocation to the LEOSs is proposed to minimize the total transmission energy. We also consider the power allocation to minimize the total transmission time given the total transmission energy. For the LEOS-GEOS uplink, a flexible partial-downloading coding transmission scheme is proposed to guarantee data reconstructability, where the joint uploaded-data size and power allocations are proposed to minimize the total transmission energy or the total transmission time. Extensive simulation results are presented to evaluate the proposed algorithms and show that regenerating code can achieve lower transmission energy and shorter transmission time for data regeneration than conventional maximum-distance separable (MDS) code.

cs.IT↗

arXiv2016

Machine-to-Machine (M2M) Communications in Software-defined and Virtualized Cellular Networks

Machine-to-machine (M2M) communications have attracted great attention from both academia and industry. In this paper, with recent advances in wireless network virtualization and software-defined networking (SDN), we propose a novel framework for M2M communications in software-defined cellular networks with wireless network virtualization. In the proposed framework, according to different functions and quality of service (QoS) requirements of machine-type communication devices (MTCDs), a hypervisor enables the virtualization of the physical M2M network, which is abstracted and sliced into multiple virtual M2M networks. In addition, we develop a decision-theoretic approach to optimize the random access process of M2M communications. Furthermore, we develop a feedback and control loop to dynamically adjust the number of resource blocks (RBs) that are used in the random access phase in a virtual M2M network by the SDN controller. Extensive simulation results with different system parameters are presented to show the performance of the proposed scheme.

cs.NI↗

arXiv2016

Content-Centric and Software-Defined Networking with Big Data

Many communities have researched the application of novel network architectures such as Content-Centric Networking (CCN) and Software-Defined Networking (SDN) to build the future Internet. Another emerging technology which is big data analysis has also won lots of attentions from academia to industry. Many splendid researches have been done on CCN, SDN, and big data, which all have addressed separately in the traditional literature. In this paper, we propose a novel network paradigm to jointly consider CCN, SDN, and big data, and provide the architecture internal data flow, big data processing and use cases which indicate the benefits and applicability. Simulation results are exhibited to show the potential benefits relating to the proposed network paradigm. We refer to this novel paradigm as Data-Driven Networking (DDN).

cs.NI↗