In multivariate long-term time series forecasting, the success of self-attention is commonly attributed to the attention matrix that encodes token interactions.In this paper, we provide evidence that challenges this view.Through extensive experiments on three classic and three latest Transformer models, we find that dotproduct attention can be replaced by elementwise operations without token interaction, such as the addition and Hadamard product, while maintaining or even improving accuracy.This motivates our central hypothesis: the effectiveness of self-attention in this task arises not from the dynamic attention matrix, but from the multi-branch feature extraction enabled by the parallel Query, Key, and Value projections and their fusion.To validate this hypothesis, we construct a minimalist multi-branch MLP that isolates the 'multi-branch mapping with element-wise operation' structure from the Transformer and show that it achieves competitive performance.Our findings indicate that the source of performance in self-attention is often misinterpreted, as its actual advantage stems from the architectural principle of multi-branch mapping and fusion, rather than the attention matrix.

Deep Packet Inspection (DPI) has been extensively employed for network security. It examines traffic payloads by searching for regular expressions (regex) with the Deterministic Finite Automaton (DFA) model. However, as the network bandwidth and ruleset size are increasing rapidly, the conventional DFA model has emerged as a significant performance bottleneck of DPI. Leveraging the Single-Instruction-Multiple-Data (SIMD) instruction to perform state transitions can substantially boost the efficiency of the DFA model. In this paper, we propose Hyperflex, a novel SIMD-based DFA model designed for high-performance regex matching. Hyperflex incorporates a region detection algorithm to identify regions suitable for acceleration by SIMD instructions across the whole DFA graph. Also, we design a hybrid state transition algorithm that enables state transition in both SIMD-accelerated and normal regions, and ensures seamless state transition across the two types of regions. We have implemented Hyperflex on the commodity CPU and evaluated it with real network traffic and DPI regexes. Our evaluation results indicate that Hyperflex reaches a throughput of 8.89Gbit/s, representing an improvement of up to 2.27 times over Mcclellan, the default DFA model of the prominent multi-pattern regex matching engine Hyperscan. As a result, Hyperflex has been successfully deployed in Hyperscan, significantly enhancing its performance.

Deep Packet Inspection (DPI) has been one of the most significant network security techniques. It is widely used to identify and classify network traffic in various applications such as web application firewall and intrusion detection. Different from traditional packet filtering that only examines packet headers, DPI detects payloads as well by comparing them with an existing signature database. The literal matching engine, which plays a key role in DPI, is the primary determinant of the system performance. FDR, an engine that utilizes 3 SIMD operations to match 1 character with multiple literals, has been developed and is currently one of the fastest literal matching engines. However, FDR has significant performance drop-off when faced with small-scale literal rule sets, whose proportion is more than 90% in modern databases. In this paper, we designed Teddy, an engine that is highly optimized for small-scale literal rule sets. Compared with FDR, Teddy significantly improves the matching efficiency by a novel shift-or matching algorithm that can simultaneously match up to 64 characters with only 15 SIMD operations. We evaluate Teddy with real-world traffic and rule sets. Experimental results show that its performance is up to 43.07x that of Aho-corasick (AC) and 2.17x that of FDR. Teddy has been successfully integrated into Hyperscan, together with which it is widely deployed in modern popular DPI applications such as Snort and Suricata.

The rapid deployment of low Earth orbit (LEO) satellite systems has propelled various space-based applications (e.g., agricultural monitoring and disaster response), which increasingly rely on advancements in deep learning (DL). However, ground stations (GS) cannot download such massive raw data for centralized training due to intermittent connectivity between satellites and GS, while the scaled-up DL models pose substantial barriers to distributed training on resource-constrained satellites. Although split learning (SL) has emerged as a promising solution to offload major training workloads to GS via model partitioning while retaining raw data on satellites, limited satellite-GS connectivity and heterogeneity of satellite resources remain substantial barriers. In this paper, we propose S-Leon, an SL framework tailored to tackle these challenges within heterogeneous LEO satellite networks. We develop a satellite early-exit model to eliminate training disruptions during non-contact periods and employ online knowledge distillation to incorporate ground knowledge, further enhancing satellite local training. Moreover, we devise a satellite model customization method that simultaneously accommodates the heterogeneous computation and communication capabilities of individual satellites. Lastly, we develop a partial model-agnostic training strategy to optimize the collaborative training effectiveness across customized satellite models. Extensive experiments with real-world LEO satellite networks demonstrate that S-Leon outperforms state-of-the-art benchmarks.

Time series forecasting is always important, with multivariate long-term series forecasting being its most challenging task. Here, the existing methods typically learn only in a single domain and focus on optimizing model structures, leading to incomplete information mining and imprecise predictions. To address this, we propose a generalized Mixture of Multi-Domain Experts (MoME) for multivariate long-term series forecasting. Unlike most existing methods, MoME focuses on multi-perspective information mining and fusing. To this end, MoME transforms time series into the frequency and spatial domains to learn their respective representations. MoME regards variates information as embedded features and applies fast Fourier transform to the time dimension. Then it learns embedded features in the frequency domain. In spatial domain learning, MoME applies self-attention mechanism on the variates dimension to efficiently capture dependencies among multiple variates. Finally, MoME fuses the outputs from all domains, reinterprets and integrates information across multiple domains, and predicts future time series. Extensive experiments prove that MoME outperforms state-of-the-art (SOTA) methods. Code is available at: https://github.com/lxy-PhD2022/MoME

Clustered federated learning (CFL) addresses the performance challenges posed by data heterogeneity in federated learning (FL) by organizing edge devices with similar data distributions into clusters, enabling collaborative model training tailored to each group. However, existing CFL approaches strictly limit knowledge sharing to within clusters, lacking the integration of global knowledge with intra-cluster training, which leads to suboptimal performance. Moreover, traditional clustering methods incur significant computational overhead, especially as the number of edge devices increases. In this paper, we propose LCFed, an efficient CFL framework to combat these challenges. By leveraging model partitioning and adopting distinct aggregation strategies for each sub-model, LCFed effectively incorporates global knowledge into intra-cluster co-training, achieving optimal training performance. Additionally, LCFed customizes a computationally efficient model similarity measurement method based on low-rank models, enabling real-time cluster updates with minimal computational overhead. Extensive experiments show that LCFed outperforms state-of-the-art benchmarks in both test accuracy and clustering computational efficiency.

Spiking Neural Network (SNN) simulation involves emulating the activation and firing of spiking neurons on hardware platforms. This is a highly time-sensitive task, requiring the simulation of billions of neurons and their intercommunication within a few milliseconds. Each neuron performs a complex, interdependent multi-stage communication and computation task. We consider the task placement of SNN on supercomputers to accelerate SNN simulation. Existing task placement methods for SNN simulations have two major limitations. First, they lack the capability to handle large-scale SNNs with billions of neurons. Second, they focus primarily on optimizing communication delay, while neglecting multi-stage computation delays in SNN simulations. In this paper, we formalize the SNN Joint Multi-stage Communication and Computation Placement (SJCCP) problem. We demonstrate that SJCCP can be solved using an approximation algorithm with an approximation ratio of $O\left( {{k^2}\sqrt {\log n\log k} } \right)$, where n is the number of voxels in the SNN and k is the number of GPUs. To further reduce the time complexity of solving SJCCP in practice, we propose a novel efficient framework, FastSJP, tailored for large-scale SNN placement. Then we apply the FastSJP framework to a human brain simulation that runs a large-scale SNN model derived from authentic biological data on a supercomputer equipped with 1024 GPUs. Experimental results verify that our framework notably reduces time overhead, ranging from 17.31% to 28.45%, compared to state-of-the-art methods. Leveraging the computational power of the supercomputer, FastSJP maximizes the problem size and processing performance, significantly advancing the development of brain-inspired intelligence.

Multi-scale analysis, a classical and crucial methodology, is extensively employed in multivariate long-term time series forecasting. However, the existing methods struggle to model arbitrary scales, thus limiting their ability to delve into complex patterns, which in turn limits the forecasting precision. Therefore, we propose the HMSformer, a multi-scale Transformer that includes a shifted stacked embedding mechanism and a unified hierarchical framework, solving this problem in terms of the in-depth and comprehensiveness of multi-scale analysis. The mechanism conducts multi-scale modeling by sampling at different positions and sizes. This sampling reflects the arbitrariness of scales and allows for a thorough scan of any scale, supporting a deeper analysis of temporal relationships. Additionally, it converts input series into structured embeddings optimized for Transformer input, allowing the self-attention mechanism to effectively learn intricate cross-scale relationships. Moreover, the unified hierarchical framework integrates multi-scale modeling across different levels of abstraction by unifying global structures, local segments, and fine details into a consistent representation. Together, these two innovations ensure that HMSformer achieves a more comprehensive and in-depth analysis of temporal relationships. HMSformer demonstrates its effectiveness with a significant improvement, achieving an average mean squared error (MSE) that is 9.925% lower than the baseline across nine authoritative datasets. Source code is available at: https://github.com/lxy-PhD2022/HMSformer

Facing the key challenge in multivariate long-term series prediction, namely effectively modeling the long-term dependencies among high-dimensional variables, we propose an innovative forest network (ForeNet). Firstly, we construct a polytree, which takes variable as leaf node, convolution as edge, and progressive fusion as the root node. Polytree models the dependencies among variables from the bottom up, adopting a local-to-global progressive learning strategy. Moreover, polytree embeds the entire sequence as input channels for convolution, allowing interactions across arbitrary time steps between variables. Thereby, polytree could model long-term dependencies. Then, we construct multiple polytrees with varying branching factors, utilizing the self-attention mechanism to assign weights and combine polytrees into a forest. Forest adopts an ensemble learning strategy to capture complex patterns hidden under high-dimensional variables. Combining progressive and ensemble strategies, ForeNet could effectively address the key challenge mentioned at the beginning. Extensive experiments show that ForeNet could reduce the average MSE by up to 11.53% compared to the baseline, which verifies the effectiveness of ForeNet. Source code is available at: https://github.com/lxy-PhD2022/ForeNet

In adversarial and untrusted network environments characterized by strict traffic controls and infrastructure surveillance, communications face persistent threats to reliability, security, and accessibility. To counter these challenges, we present the Generative Resilient Network (GRN), a three-layer architecture that combines cloud-based resource agility, hybrid secure transmission, and large language model (LLM)-driven adaptive control to maintain connectivity under adversarial conditions. Preliminary experiments demonstrate GRN’s scalability, robustness, and resilience against interference across both low-latency and high-anonymity scenarios.

Recently, large vision-language models (LVLMs) unleash powerful analysis capabilities for low Earth orbit (LEO) satellite Earth observation images in the data center. However, fast satellite motion, brief satellite-ground station (GS) contact windows, and large size of the images pose a data download challenge. To enable near real-time Earth observation applications (e.g., disaster and extreme weather monitoring), we should explore how to deploy LVLM in LEO satellite networks, and design SpaceVerse, an efficient satellite-ground synergistic LVLM inference system. To this end, firstly, we deploy compact LVLMs on satellites for lightweight tasks, whereas regular LVLMs operate on GSs to handle computationally intensive tasks. Then, we propose a computing and communication co-design framework comprised of a progressive confidence network, and an attention-based multi-scale preprocessing, used to identify on-satellite inferring data, and reduce data redundancy before satellite-GS transmission, separately. We implement, and evaluate SpaceVerse on real-world LEO satellite constellations and datasets, achieving a 31.2% average gain in accuracy and a 51.2% reduction in latency compared to state-of-the-art baselines.

Network policy plays a crucial role in cloud-native networking, especially in multi-tenant scenarios. It provides precise control over connectivity by specifying source and destination endpoints, traffic types, and other criteria to allow or deny traffic. However, manual configuration of these policies introduces the risk of errors, leading to isolation violations or network service unavailability. Therefore, network policy verification is essential for maintaining security and quality of service in cloud-native networking. Currently, a naïve approach involves individually checking each policy within the cluster, which can take over 100s for verification in a cluster size of over 100k. Existing verification frameworks, like Kano and Verikube, improve performance by leveraging pre-filtering and Satisfiability Modulo Theories (SMT) solvers, achieving a 3.12x to 12.99x performance boost over the naïve baseline. However, as network policy changes rapidly within 100ms in real cloud-native networks, both frameworks need over 10s to perform verification for cluster sizes over 100k, which is far from satisfying. To overcome these issues, we propose and implement a novel network policy verification framework NPV, which utilizes the policy-label pre-filter process with bitwise compression. We further enhance the policy verification algorithm with a policy-namespace divide-and-conquer strategy to improve the data-level parallelism. We implement NPV on commodity servers and evaluate its performance using real network policy datasets. Our experiments indicate that, compared with the state-of-the-art methods, NPV can achieve up to 139.00x to 651.06x improvement in verification time compared to Kano and Verikube, with 65% less memory usage.

Federated Learning, as an emerging edge artificial intelligence paradigm, enables a group of clients to collaboratively train a global model without revealing their local data. The conventional FL algorithms usually depend on the access to exact gradient or Hessian matrix, which may be inaccessible due to resource limitation or application restriction. Meanwhile, Federated Learning intrinsically suffers from data heterogeneity, which restricts the global model from performing well on each clients’ task. To simultaneously tackle these two challenges, we propose a gradient free personalized federated learning framework, namely pFedZO. We utilize infimal convolution to bridging the gap between personal and global knowledge, and exploit zeroth-order gradient estimator to solve the problem. We theoretically show the local approximation can converge sublinearly and the global problem converge to the neighbourhood of the optimal with a same speed. We further propose pFedZO-Heur to accelerate training procedure. Experimentally, we verify that pFedZO excels at test accuracy with the vanilla Zeroth-Order Optimization (ZOO) based FL by <?TeX $5\%$?> Math 1 . We also show pFedZO-Heur can achieve the same performance level with lower time consumption.

Deep Packet Inspection (DPI) is a significant network security technique. It examines traffic workloads by searching for specific rules. Since every byte of packets needs to be examined by many literal rules, multi-literal matching becomes the performance bottleneck of DPI. FDR, the fastest multi-literal matching engine on CPUs, takes advantage of Single-Instruction-Multiple-Data (SIMD) to alleviate this bottleneck and achieves a performance boost over the widely-used Aho-Corasick (AC) algorithm. However, FDR does not deeply exploit the data-level parallelism of SIMD and its SIMD vector utilization is only 50%. Besides, limited by certain SIMD shift instructions, it cannot benefit from advanced SIMD instruction sets. To overcome these issues, we propose Harry, a scalable and SIMD-based multi-literal matching engine. Harry adopts a column-vector-based matching algorithm to improve the data-level parallelism and SIMD vector utilization. To support the algorithm, it takes two encoding methods to compress the mask table. Also, it utilizes shuffle instruction to implement shift. We implement Harry on commodity CPU and evaluate it with real network traffic and DPI rules. Our evaluation shows that Harry reaches a throughput of 30∼70Gbit/s, up to 52x that of AC and 2.09x of FDR. It has been successfully deployed in Hyperscan.

A Service Function Chain (SFC) is an ordered sequence of network functions (NFs). Though cost-effective, software-based NFs could introduce a significant performance penalty. In this paper, we present P4SFC, a high-performance and flexible SFC system that leverages the capability of emerging programmable switches. We seek to accelerate packet processing in SFC by offloading proper NFs to P4-capable switches. First, considering the current limitations of P4, we analyze the offloadability of NFs at different granularities in detail, and enable P4SFC to generate offloading strategies for both partially and fully offloadable SFCs. Second, to deploy new SFCs at runtime, we design a dynamic P4 data plane, of which the execution logic can be reconfigured at runtime without interrupting the existed execution logic. Third, to efficiently utilize the limited memory in programmable switches, we propose a state allocator to dynamically offload those NF states that bring the highest performance profits according to the recent flow distribution. We demonstrate the feasibility and practicality of P4SFC with our implementation on a commodity Tofino-based programmable switch. Experimental results show that P4SFC achieves significant performance improvement for real SFC implementations.

With the aging of the population, caregiving for the elderly has become an urgent social topic. While the rapid development of data-driven technologies provides tremendous promises to deal with this issue, the gap between data-driven and practical healthcare challenges their effectiveness. With the aim of providing strong data basis for the large-scale data-driven healthcare, this paper proposes a human-space-service based method, named CareMap, to model and quantity the caregiving process for the elderly aging at home. We build the intelligent individual profile to model and compute seniors’ health conditions on the one hand and care network profile to model and quantify the care resources around the elderly on the other hand. We design a CareMap based prototype to illustrate the possible application and discuss its potentials and limitations.

Using data-driven technologies to support the healthcare of the elderly has been largely celebrated as an effective means. This paper focuses on the issue of using video-based sensing technologies to remotely monitor the activities and conditions of the elderly. Although it is a widely explored field, the high cost and high infrastructural requirements of most existing technologies usually challenge their effectiveness and efficiency in practical caregiving context. To address these challenges, we propose a lightweight network based real-time anomaly detection system, which consists of video-based ADL sensing and pre-processing, AI streaming aggregating and cluster computing. We examine our method by implementing and deploying it into a real-world care facility for the elderly in Shanghai China. The results show that our method has good performance in expansibility, reliability, bandwidth availability, accuracy and privacy protection.

Using data-driven technologies to support healthcare has been largely celebrated as an effective means. Yet, most existing data-driven technologies are designed with the purposes of improving productivity, efficiency, and effectiveness etc., i.e. care administer-centric design. The needs and perceptions of caregiver, the actual users of most technologies, are largely ignored. In this paper, we examine the impacts of data-driven technologies from the perspective of caregivers. Through a questionnaire study with 191 caregivers in Shanghai China, we quantify the impacts of data-driven technologies to caregivers' work and experiences. Our results show that while the embedded data-driven technologies provide significant benefits to caregivers' work, these benefits are often limited by the complex, situated and fragmented nature of healthcare. We analyze the results and propose our suggestions to the further design.

Message queues are fundamental components in modern cloud architecture. With the increased data traffic, traditional distributed message queue systems suffer from low performance. We present NetMQ, a new message queue architecture that leverages the power and flexibility of new-generation programmable switches to handle message producing and consuming requests of hot message queues. Due to the limited switch memory, we design a memory layout for multiple cached queues sharing a single register array, and two memory supplementing strategies for different types of queues. To handle dynamic workloads, we design a space-efficient hot queue detector in the data plane, and a heuristic algorithm running in the controller to update the cached topic partition regularly. We implement a NetMQ prototype on Barefoot Tofino switches and commodity servers. Our evaluations show that NetMQ reduces the latency by 3.5-18×, and improves the throughput by 3-3.8× for queues cached in the switch, while incurs negligible overheads for the uncached queues.

Network Function Virtualization (NFV) has inspired numerous orchestration algorithms to decide Virtualized Network Function (VNF) placement and routing paths for service requests with Service Function Chain (SFC) demands. With different optimization goals, these algorithms may select various routing paths for request flows. Nevertheless, existing traffic steering solutions either fail to fully support NFV Orchestrator (NFVO) in path selection, or result in low scalability in the underlay Software Defined Network (SDN). In this paper, we propose STAR to tackle both problems simultaneously. STAR divides the entire routing path for SFC request into several path segments, then performs Output-Port-based, Default-Path-based or RSP-ID-based traffic steering on each path segment. With the idea of path division and these traffic steering mechanisms, STAR achieves flexible traffic steering along any paths selected by NFVO and enables different Rendered Service Paths (RSPs) with the same path segments to share the same forwarding rules. In this way, STAR ensures the correct enforcement of path selection decisions from NFVO and significantly reduces forwarding rule consumption and control overhead. We evaluate STAR with the experiments on a real testbed and large-scale simulations. The results show that our framework is scalable in the SDN data plane and control plane (e.g., reducing rules in hardware switches by more than 70% compared with NSH-based solutions) while retaining the flexibility of steering traffic along any SFC routing paths to provide full support for path decision enforcement (e.g., achieving the highest path decision enforcement ratio in 11 of the 12 cases) with acceptable overhead.

The Internet of Things (IoT) technology provisions unprecedented opportunities to evolve the interconnection among human beings. However, the latency brought by unstable wireless networks and computation failures caused by limited resources on IoT devices prevents users from experiencing high efficiency and seamless user experience. To address these shortcomings, the integrated Mobile Edge Computing (MEC) with remote clouds is a promising platform to enable delay-sensitive service provisioning for IoT applications, where edge-clouds (cloudlets) are co-located with wireless access points in the proximity of IoT devices. Thus, computation-intensive and sensing data from IoT devices can be offloaded to the MEC network immediately for processing, and the service response latency can be significantly reduced. In this paper, we first formulate two novel optimization problems for delay-sensitive IoT applications, i.e., the total utility maximization problems under both static and dynamic offloading task request settings, with the aim to maximize the accumulative user satisfaction on the use of the services provided by the MEC, and show the NP-hardness of the defined problems. We then devise efficient approximation and online algorithms with provable performance guarantees for the problems in a special case where the bandwidth capacity constraint is negligible. We also develop efficient heuristic algorithms for the problems with the bandwidth capacity constraint. We finally evaluate the performance of the proposed algorithms through experimental simulations. Experimental results demonstrate that the proposed algorithms are promising in reducing service delays and enhancing user satisfaction, and the proposed algorithms outperform their counterparts by at least 10.8 percent.

Packet routing is one of the fundamental problems in computer networks in which a router determines the next-hop of each packet in the queue to get it as quickly as possible to its destination. Reinforcement learning (RL) has been introduced to design autonomous packet routing policies with local information of stochastic packet arrival and service. However, the curse of dimensionality of RL prohibits the more comprehensive representation of dynamic network states, thus limiting its potential benefit. In this article, we propose a novel packet routing framework based onmultiagentdeep RL (DRL) in which each router possess anindependentlong short term memory (LSTM) recurrent neural network (RNN) for training and decision making in afully distributedenvironment. The LSTM RNN extracts routing features from rich information regarding backlogged packets and past actions, and effectively approximates the value function of Q-learning. We further allow each route to communicate periodically with direct neighbors so that a broader view of network state can be incorporated. The experimental results manifest that our multiagent DRL policy can strike the delicate balance between congestion-aware and shortest routes, and significantly reduce the packet delivery time in general network topologies compared with its counterparts.

Chronic gastritis (CG) is a highly prevalent disease of the digestive system. In the diagnosis model of traditional Chinese medicine (TCM), whether a patient has a certain syndrome is determined through a combination of symptoms. However, TCM diagnosis for CG syndromes has proven quite challenging. First, due to the large number of symptoms, the correctness of diagnosis largely depends on the doctor’s experience, which might be subjective. Second, collecting all the symptoms for diagnosis in advance is time-consuming.To address the two challenges, we first design an ensemble learning model for the diagnosis of CG syndrome using LightGBM (Light Gradient Boosting Machine). The model can diagnose new CG syndrome instances effectively. We also adopt a voting mechanism with four feature selection algorithms to select the most relevant symptoms. We collected a total of 2,680 CG cases, of which 536 (20%) cases were used as a test set to evaluate the proposed model. The model with LightGBM has an average diagnosis accuracy of 91.64% for 10 syndromes, which is higher than two prevailing methods kNN and SVM. In addition, we also identified 36 most relevant symptoms for the diagnosis of Piwei Shire pattern with four feature selection algorithms. The diagnosis accuracy obtained with partial features can reach 86.75%, which is slightly higher than that of full-feature diagnosis. The results reveal that using only partial features that have the most significant impact on the diagnosis results will not reduce the diagnosis accuracy.

By decoupling network functions from the underlying dedicated hardware, network function virtualization (NFV) has become a promising paradigm to reduce network operating expenses. NFV can provide elastic placement of Virtual Network Functions (VNFs) in the underlying data centers. However, the co-located VNFs on the same server may suffer from performance interference due to computing-resource and memory-resource sharing. This article focuses on how to ensure the performance of each VNF while minimizing the total energy consumption of the data center. By showing that the bin-packing problem is polynomial-time reducible to our model, we prove that the offline version of this problem is NP-complete. Then, for a homogeneous environment where all servers are of the same type, we design First-Fit Heuristic (FFH) algorithm and analyze the approximation performance of it by proving the lower bound value. For the heterogeneous environments, we propose an efficient solution based on deep reinforcement learning (DRL) named DDAP (Deep Deterministic Automatic Placement). Our experiments show that DDAP can quickly respond to each request and achieve better performance. In particular, DDAP can reduce energy consumption by 7.6% and running time cost by 63.2% on average compared to state-of-the-art methods.

Owing to the ubiquitous penetration of Wi-Fi in our daily lives, Wi-Fi indoor localization has attracted intensive attentions in the last decade or so. Despite some significant progresses, the high accuracy of existing systems is still achieved at the cost of dense access point (AP) deployment. The more practical single AP localization is largely left as an open problem because the hardware-induced time delay “contaminates” the measurement of signal propagation time in the air. In this article, we design and implement M3to tackle this challenge with commodity Wi-Fi cards. M3exploits a multipath-assisted approach that turns the harmful multipath from foe to friend to enable single AP localization: a device can be pinpointed through the combination of azimuths and the relative time of flight (ToF) of Line-of-Sight (LoS) signal and reflection signals, eliminating the need for multiple APs along with their absolute ToF measurements. M3further utilizes frequency hopping to combine multiple channels to form a virtually wider-spectrum channel for higher ToF resolution. As a prominent feature of M3, the channels do not need to be adjacent. Comprehensive experiments demonstrate that M3outperforms the state-of-the-art systems and achieves a median localization accuracy of 71 cm in three environments with a single AP.

Since Network Function Virtualization (NFV) decouples network functions (NFs) from the underlying dedicated hardware and realizes them in the form of software called Virtual Network Functions (VNFs), they are enabled to run in any resource-sufficient virtual machines. A service function chain (SFC) is composed of a sequential set of VNFs. As VNFs are vulnerable to various faults such as software failures, we consider how to deploy both active and standby SFC instances. Given the complexity and unpredictability of the network state, we propose a double deep Q-networks based online SFC placement scheme DDQP. Specifically, DDQP uses deep neural networks to deal with large continuous network state space. In the case of stateful VNFs, we offer constant generated state updates from active instances to standby instances to guarantee seamless redirection after failures. With the goal of balancing the waste of resources and ensuring service reliability, we introduce five progressive schemes of resource reservations to meet different customer needs. Our experimental results demonstrate that DDQP responds rapidly to arriving requests and reaches near-optimal performance. Specifically, DDQP outweighs the state-of-the-art method by 16.30% and 38.51% higher acceptance ratio under different schemes with 82x speedup on average. In order to enhance the integrity of the SFC state transition, we further proposed DDQP+, which extends DDQP by adding the delayed placement mechanism. Compared with DDQP, the design of the DDQP+ algorithm is more reasonable and comprehensive. The experiment results also show that DDQP+ achieved further improvement in multiple performance indicators.

A Service Function Chain (SFC) is an ordered sequence of network functions (NFs). With the emerging Network Function Virtualization (NFV) paradigm, NFs can be deployed as software instances on commodity servers, leading to a more flexible provision of network service. However, the flexibility of NFV comes with considerable compromises since virtual network functions (VNFs) introduce significant processing latency overheads. In this article, we design a flexible SFC parallel system called FlexChain, enabling the parallelism among VNFs to reduce the processing latency of SFCs. To leverage the benefits of parallelism, we study the problem of joint optimization over SFC parallelism and placement with the objective of accepting as many requests as possible. Since the problem is proved to be NP-hard, we propose a parallelism-aware approximation placement algorithm with performance guarantees, and an efficient heuristic algorithm for large-scale data center networks. Our simulation results show that FlexChain combined with our placement algorithm can substantially improve the number of accepted flows in the latency-sensitive scenario, and significantly reduce the average latency of accepted flows at the same time.

Programmable data plane hardware creates a possibility to solve network-related problems. Ensuring fault tolerance of link failures is a fundamental network issue. Link failure recovery mechanisms are widely used in traditional and software-defined networks. The proactive failure recovery mechanism usually requires a backup path to be installed in the switch in advance. When the link fails, the switch can quickly switch to the backup path to continue sending packets. However, storing a large number of backup paths consumes a lot of switch storage. In this article, we analyze why implementing traditional proactive failure recovery mechanism introduces huge switch storage overhead, and discuss the flexibility and limitations of the programmable data plane. Then, we present P4Neighbor, a proactive link failure recovery framework based on the programmable data plane. P4Neighbor encapsulates backup paths into the header of a packet when a link failed and leverages this information to achieve link failure recovery. By storing only the backup paths of the neighbor switches, P4Neighbor requires little switch storage to store backup paths. Besides, P4Neighbor also takes complex link failure situations into consideration, which makes the network's fault tolerance slightly increase. Experimental results show that compared with the traditional failure recovery mechanism, P4Neighbor achieves a reduction rate of 57.9%-84.5% in terms of stored switch entries. Meanwhile, P4Neighbor also has a higher failure recovery ratio than traditional proactive link failure recovery mechanisms.

The dynamic nature of software-defined networking requires frequent updates to the flow table in the data plane of switches. Therefore, the ternary content-addressable memory (TCAM) used in switches to match packet header fields against forwarding rules needs to support high rates of updates. Existing off-the-shelf switches update rules in batches for efficiency, but may suffer from forwarding inconsistencies during the batch update. In this paper, we design and evaluate a TCAM update optimization framework that can guarantee consistent forwarding during the entire update process while making use of a layered TCAM structure. Our approach is based on a modified-entry-first write-back strategy that significantly reduces the overhead from movements of TCAM entries. In addition, our approach detects reordering cases, which are handled using efficient solutions. Based on our evaluation results, we can reduce the cost of TCAM updates by 30%-88% compared to state-of-the-art techniques.

A Service Function Chain (SFC) is an ordered sequence of network functions (NFs). Software-based NFs in Network Function Virtualization (NFV) could introduce significant performance overhead. In this paper, we present P4SFC, a high-performance SFC system that leverages P4-capable switches to accelerate packet processing by offloading proper NFs to the switches. First, considering the current limitations of P4, we analyze the offloadability of NFs and divide them into three categories: fully offloadable, partially offloadable, and non-offloadable. Second, when deploying new SFCs, P4SFC automatically offloads proper NFs to switches based on their position and offloadability. To deploy new SFCs at runtime, we design a dynamic P4 data plane, whose execution logic can be reconfigured at runtime without interrupting the current execution logic. Finally, to maintain state consistency between the server and the switch for partially offloaded NFs, we design a state library to automatically synchronize states between servers and switches. Experimental results show that P4SFC achieves significant performance improvement for real-world SFCs.

Wildcard rules are implemented in various important networking scenarios, including QoS, firewall, access control, and network traffic monitoring and analysis. However, there are cross-rule dependencies between wildcard rules, which both increase significant overhead and affect the semantic correctness of packet classification when caching rules. Considerable efforts have been made to mitigate the impacts of the dependency issue in rule caching, but it is still a bottleneck for cache systems. In this paper, we show how to give applications the flexibility of completely dependency-free wildcard rule caching by decoupling the cached rules and their dependent rules. Our FreeCache scheme has wide applicability to packet classification devices with wildcard rule caching. We validate the effectiveness of FreeCache through two respects: (1) Implementing various cache algorithms (e.g., LSTM) and cache replacement algorithms (e.g., ARC, LIRS) that are difficult to use in dependency-bound situations in the cache system with FreeCache. (2) Developing a prototype in a Software-Defined Network (SDN), where hybrid OpenFlow switches use TCAM as cache and RAM as auxiliary memory. Our experimental results reveal that FreeCache improves the cache performance by up to 60.88% in the offline scenario. FreeCache also offers the promise of applying any existing caching algorithms to wildcard rule caching while guaranteeing the properties of semantic correctness and equivalence.

The Internet of Things (IoT) technology offers unprecedented opportunities to interconnect human beings. However, the latency brought by unstable wireless networks and computation failures caused by limited resources on IoT devices prevents users from experiencing high efficiency and seamless user experience. To address these shortcomings, the integrated MEC with remote clouds is a promising platform, where edge-clouds (cloudlet) are co-located with wireless access points in the proximity of IoT devices, thus intensive-computation and sensing data from IoT devices can be offloaded to the MEC network for processing, and the service response latency can be significantly reduced. In this paper, we study delay-sensitive service provisioning in an MEC network for IoT applications. We first formulate two novel optimization problems, i.e., the total utility maximization problems under both static and dynamic offloading task request settings, with the aim to maximize the accumulative user satisfaction of using the services provided by the MEC. We then show that the defined problems are NP-hard. We instead devise efficient approximation and online algorithms with provable performance guarantees for the problems. We finally evaluate the performance of the proposed algorithms through experimental simulations. Experimental results demonstrate that the proposed algorithms are promising.

Software Defined Networking (SDN) enables flexible flow control by deploying fine-grained rules in OpenFlow switches. Modern commodity switches usually use TCAM to store these rules and perform high-speed parallel lookups. Though efficient, the TCAM capacity is limited because TCAM is expensive in cost and power-hungry. The explosive growth in the number of rules has exacerbated the limitation of TCAM. There have been considerable efforts in implementing hybrid flow tables with both TCAM and RAM, where the high-speed TCAM is regarded as a cache to store the most popular rules and the cheap RAM is used to handle cache miss. The primary challenges for designing hybrid TCAM/RAM flow tables lie in how to improve cache hit rate and how to handle wildcard rule dependency when allocating rules between TCAM and RAM.

Packet routing is one of the fundamental problems in computer networks in which a router determines the next-hop of each packet in the queue to get it as quickly as possible to its destination. Reinforcement learning has been introduced to design the autonomous packet routing policy namely Q-routing only using local information available to each router. However, the curse of dimensionality of Q-routing prohibits the more comprehensive representation of dynamic network states, thus limiting the potential benefit of reinforcement learning. Inspired by recent success of deep reinforcement learning (DRL), we embed deep neural networks in multi-agent Q-routing. Each router possesses an independent neural network that is trained without communicating with its neighbors and makes decision locally. Two multi-agent DRL-enabled routing algorithms are proposed: one simply replaces Q-table of vanilla Q-routing by a deep neural network, and the other further employs extra information including the past actions and the destinations of non-head of line packets. Our simulation manifests that the direct substitution of Q-table by a deep neural network may not yield minimal delivery delays because the neural network does not learn more from the same input. When more information is utilized, adaptive routing policy can converge and significantly reduce the packet delivery time.

With an increasing demand for flexible management in software-defined networks (SDNs), it becomes critical to minimize the network policy update time. Although major SDN controllers are now optimized for rapid network update at the control plane, there is still room for data plane optimization in terms of update time, when using TCAM-based physical SDN commodity-off-the-shelf switches. A slow update directly affects network performance and creates bottlenecks. To minimize the flow entry update time, a dependency graph, a kind of directed acyclic graph (DAG), can be used for the access management of flow entries at the switch. Thanks to the DAG, unnecessary entry movements, which are the main factor slowing down flow entry updates, can be avoided. However, existing algorithms show limitations when updates become very frequent. We propose a new flow entry update algorithm, called FastRule, that exploits a greedy strategy with an efficient data structure to accelerate flow entry update with a DAG approach. Moreover, we also adjust our algorithm for other flow table layouts to make it scalable. We elaborate on the correctness of FastRule and test our algorithm using a hardware switch. Compared with existing algorithms, the evaluation shows that our algorithm is about 100x faster than state-of-the-art solutions with a flow table of 1k size.

With an increasing demand for resilience in software-defined networks (SDN), it becomes critical to minimize service recovery delay upon route failures. Fast reroute (FRR) mechanisms are widely used in IP and MPLS networks by computing the recovery path before a failure occurs. The centralized control plane in SDN can potentially enhance path computation, so that FRR path computation can better scale in SDN than in traditional networks. However, the traditional FRR path computation algorithms could lead to a poor performance in large-scale SDN. The problem can become more severe for a highly dynamic network, which often sees dozens of failures or configuration changes in any single day. We propose a new algorithm that exploits pruned searching to quickly compute recovery paths for all-pair switches/hosts upon a link failure. For applications requiring stringent path robustness levels, we also extend this algorithm to quickly find the shortest guaranteed-cost path, which ensures that the recovery path used upon on-path link failures has the minimum cost. Compared with traditional solutions, our evaluations show that our algorithm is about 8 ~ 81 times faster than the practical implementation, 1.93 ~ 3.11 times faster than the state-of-the-art solution. Our results also show that the shortest guaranteed-cost path can reduce the cost of the recovery path significantly. Moreover, we design a prototype to show how to deploy our algorithm in an OpenFlow network.

Software-defined networking (SDN) provides flexible network control, which has enabled new, more complex network control mechanisms. These approaches impose high demands on flow updates in OpenFlow switches. Existing SDN controllers and firmware have been optimized for high-speed network updates, but specific switch implementations differ in their behavior. Previous optimization schemes for updates ignore the diversity of instruction types and switch behavior. In this paper, we present RuleTailor, an efficient, measurement-based optimization framework for SDN flow updates, to overcome these limitations. In contrast to other measurement-based optimization frameworks, RuleTailor uses new techniques, such as instruction type transformation, pseudo deletion, and match field distance, which can work in both the control plane and the data plane. Our evaluation shows that RuleTailor achieves a performance of fewer than 12 milliseconds per update in a content-addressable flow table with 1k entries, outperforming the state-of-the-art measurement-based framework, Tango, by a factor of 10.

With an increasing demand for flexible management in software-defined networks (SDNs), it becomes critical to minimize the network policy update time. Although major SDN controllers are now optimized for rapid network update at the control plane, there is still room for data plane optimization in terms of update time, when using TCAM-based physical SDN commodity-off-the-shelf switches. A slow update directly affects network performance creating bottlenecks. To minimize flow entry update time, a dependency graph, a kind of DAG (directed acyclic graph), can be used for the access management of flow entries at the switch. Thanks to the DAG, unnecessary entry movements, which are the main factor slowing down flow entry updates, can be avoided. However, existing algorithms show limitations when updates become very frequent. We propose a new flow entry update algorithm, called FastRule, that exploits a greedy strategy with an efficient data structure to accelerate flow entry update with a DAG approach. Moreover, we also adjust our algorithm for other flow table layouts to make it scalable. We elaborate on the correctness of FastRule and test our algorithm using a hardware switch. Compared with existing algorithms, the evaluation shows that our algorithm is about 100x faster than state-of-the-art solutions with a flow table of 1k line size.

Determining the shortest-path distance between vertices in the weighted graph is an important problem for a broad range of fields, such as context-aware search and route selection. While many efficient methods for querying shortest-path distance have been proposed, they are poorly suited for parallel architectures, such as multi-core CPUs or computer clusters, due to the strong task dependencies. In this paper, we propose ParaPLL, a new parallelism-friendly framework for fast shortest-path distance query on large-scale weighted graphs. ParaPLL exploits intra-node and inter-node parallelism by using shared memory and message passing paradigms respectively. We also design task assignment and synchronization policies, which allow ParaPLL to reach remarkable speedups compared to state-of-the-art solutions. Moreover, we also prove the correctness of ParaPLL. To the best of our knowledge, ParaPLL is the first parallel framework that utilizing pruned landmark labeling to accelerate shortest-path distance queries on large-scale weighted graphs. Our evaluation results show that ParaPLL is 9.46 times faster than the corresponding serial version on a weighted 0.3M-vertex graph using a 12-core computer. ParaPLL on a 6-node computer cluster can also achieve a speedup of up to 5.6 over the single-node implementation.

The fundamental tasks of the control plane in software defined networking (SDN) are to customize forwarding policies for the data plane and to provide global network view for applications. The logically centralized control plane design brings benefits in terms of network programmability and can largely ease network management. However, it also increases efficiency concerns. One practical control plane challenge is path computation, because it can require a significant amount of computation load if the network scale is large and the path requests from applications are frequent. In this paper, our goal is to build a high-performance control plane for path computation using multiple controllers. Previous works attempt to improve control plane efficiency by balancing only the load for data plane behavior between multiple controllers. Going beyond conventional wisdom, we designed ParaCon, a solution we propose to speed up the control plane by distributing the load of path computation. We also address the consistency and synchronization overhead challenges related to ParaCon design. To the best of our knowledge, ParaCon is the first attempt that utilizes node parallelism in SDN path computation. We evaluated ParaCon using both Mininet and real-world clusters. Our results show that the path computing time of ParaCon can achieve a speedup of 10× over Floyd (used in POX) and Dijkstra (used in ONOS) baseline implementations for networks with hundreds of nodes.

In the past decade, Location-Based Social Networks (LBSNs) have attracted attentions from both the academia and industry. All LBSN sites are venue-centric, i.e., each review or check-in must be associated with a venue. Despite the importance of the venues, there still lacks a systematical study on LBSN sites from the venues' perspective. To fill this gap, we conduct a comprehensive study on Dianping, the largest online review site in China, with data of more than 506K venues. We first characterize the demographics of each venue. We then measure the venue popularity through real-time reviews. Finally, we propose the concept of “venue network” to study the linkage among the venues. Our paper not only provides a clear picture of venues on Dianping, but also sheds light on potential applications in LBSNs.

Shortest-path queries on weighted graphs are an essential operation in computer networks. The performance of such algorithms has become a critical challenge in emerging software-defined networks (SDN), since SDN controllers need to perform a shortest-path query for every flow. Unlike classic solutions (e.g., Dijkstra's algorithm), high-performance shortest-path query algorithms include two stages: preprocessing and query answering. Despite the improved query answering time, existing two-stage algorithms are still extremely time-consuming in preprocessing large-scale graphs. In this paper, we propose an efficient shortest-path query algorithm, called BBQ, which reduces the running time of both stages via tree decomposition. BBQ constructs a distance oracle in a bottom-top-bottom manner, which significantly reduces preprocessing time over existing algorithms. In addition, BBQ can answer batch queries in bulk by traversing the decomposed tree instead of executing separate queries. Our experimental results show that BBQ outperforms state-of-the-art approaches by orders of magnitude for the running time of the preprocessing stage. Meanwhile, BBQ also offers remarkable acceleration for answering batches of queries. As a result, SDN controllers that use BBQ can sustain 1.1-27.9 times higher connection request rates.

Software-Defined Networking (SDN) has great potentials in changing the fragile and complex enterprise networks. One operational challenge to SDN deployment is the settlement of legacy switches. A hybrid SDN consisting of both SDN and legacy switches may be a tradeoff. Nevertheless most of the current SDN control planes can not handle legacy switches. To overcome this problem, we present HybridFlow, a lightweight control plane for hybrid SDN. HybridFlow can abstract a hybrid network into a logical SDN network and existing SDN control applications can run on it transparently.

OpenFlow has been widely used in Software Defined Networking (SDN) to customize data plane behaviors through policies given by a logically centralized controller. The centralized control plane design brings the potential of simplifying network management, but it also raises scalability concern for large-scale networks. In this paper, we aim to build a highly scalable and flexible OpenFlow control plane. We propose the design and implementation of cCluster, which leverage the parallelism of cluster to balance the control plane load. Compared with existing solutions, cCluster can achieve better scalability, and can enable elastic management. We evaluated the scalability and response time of cCluster using Mininet, and our results show that cCluster can serve thousands of flow entries simultaneously, with only slightly latency penalty.

Network coding encourages information flow mixing in a network. It helps increase the throughput and reduce the cost of data transmission, especially for one-to-many multicast applications. An interesting problem is to understand and quantify the coding advantage and cost advantage, i.e., the potential benefits of network coding, as compared to routing, in terms of increasing throughput and reducing transmission cost, respectively. Two classic network models were considered in previous studies: directed networks and undirected networks. This work further focuses on two types of parameterized networks, including bidirected networks and hyper-networks, generalizing the directed and the undirected network models, respectively. We prove upper- and lower-bounds on multicast coding advantage and cost advantage in these models.

Network coding is a technique that allows information flows to be encoded while routed across a data network. It was shown that network coding helps increase the throughput and reduce the cost of data transmission, especially for one-to-many multicast applications. An important direction in network coding research is to understand and quantify the coding advantage and cost advantage, i.e., the potential benefits of network coding, as compared to routing, in terms of increasing throughput and reducing transmission cost, respectively. Two classic network models were considered in previous studies of coding advantage: directed networks and undirected networks. The study of coding advantage in this work further focuses on two types of parameterized networks, including bidirected networks and hyper-networks, which generalizes the directed and the undirected network models, respectively. With proper parameter setting, more realistic modeling of networks in practice can be achieved. We prove upper-bounds and lower-bounds on the coding advantage for multicast in these models. Some of our bounds are new and unknown before, some improve upon previously proven bounds, and some answer open questions in the literature.

Small playback lag in live streaming is important for time-critical and interactive applications such as live stock, market updates, sports and remote education. In this paper, we present Elite addresses the playback lag problem in peer-assisted live streaming systems. Instead of deploying a large initial offset to all the users, Elite seeks the possibility of initializing users with layered proportional initial scheduling point, thus achieving differentiated playback lag service for the system. For saving server bandwidth and reducing lag time, Elite employs a novel strategy which arranges peers into a virtual tree structure and quantifies playback lag of each layer that finally converges to a constant value. This way, Elite can help users to achieve much shorter average playback lag and prioritized service within the same channel. As illustrated in our design, analysis, and simulation studies, Elite is able to fully exploit limited pool of server bandwidth to support peer-assisted live streaming with prioritized playback lag, and achieves shorter average playback lag compared with synchronized strategy, such as R2.

As the physical link speeds grow and the size of routing table continues to increase, IP address lookup has been a challenging problem at routers. There have been growing demands in achieving high-performance IP lookup cost-effectively. Existing approaches typically resort to specialized hardwares, such as TCAM. While these approaches can take advantage of hardware parallelism to achieve high-performance IP lookup, they also have the disadvantage of high cost. This paper investigates a new way to build a cost-effective IP lookup scheme using graphics processor units (GPU). Our contribution here is to design a practical architecture for high-performance IP lookup engine with GPU, and to develop efficient algorithms for routing prefix update operations such as deletion, insertion, and modification. Leveraging GPU's many-core parallelism, the proposed schemes addressed the challenges in designing IP lookup at GPU-based software routers. Our experimental results on real-world route traces show promising gains in IP lookup and update operations.

Short playback lag is preferred in many urgent and interactive scenarios such as live sports and distance education. However, measurement studies have shown that many popular P2P live streaming systems still suffer from long playback lag, say, more than 100 seconds, which makes the live streaming le

The objective of many applications with the surveillance missions in wireless sensor networks is to provide long-term monitoring of the specific environments, such as disaster-hit regions. These applications usually perform continuous monitoring without any maintenance, even if some sensor nodes fai

Graphics processing unit (GPU) has evolved into a general-purpose computing platform. Inspired by the GPU technology advantage, this paper concerns the design and performance evaluation of practical GPU-accelerated server architecture for Video-on-Demand (VoD) services with network coding. Following

It has been well known that network coding can achieve better network throughput in certain topologies by allowing coding at intermediate nodes. However, the benefit of network coding in P2P content distribution is controversial in recent literatures. In this paper, we seek to investigate how network coding performs in a BitTorrent-like protocol from a practical perspective. To validate the benefit of network coding in BitTorrent, we first implement NCTorrent, a network coding enabled BitTorrent client. We then track about 10,000 real-world BitTorrent session's user behaviors and use the user behaviors to drive the experiments. Based on the results we establish, our conclusion with respective to the benefits of network coding is pessimistic. Network coding does not help that much in real world scenarios as claimed in previous work. In particular, the logged data also reveal that the last block problem, where network coding is believed to be helpful, only appears with very low probabilities.

P2P live streaming systems suffer a lot from long playback lag in lag-sensitive scenarios. In this paper, we propose a new approach to reducing the playback lag in P2P live streaming systems. According to measurement studies, there exist a certain amount of active peers, who stay longer and contribute more bandwidth than other peers. Inspired by this, we propose a tiered overlay design, in which peers are organized into three tiers based on their degrees of activity. We develop a set of algorithms to evaluate the peers' degrees of activity. Specifically, the backbone of the overlay consists of the peers with high activity in tier-1. These active peers are responsible for diffusing the newly generated fresh chunks to peers located in all the involved Autonomous Systems (ASes). They contribute more bandwidth and thus enjoy shorter playback lag. Further more, adaptive biased neighbor selection algorithm is employed among non-backbone peers to keep traffic locality. Evaluated by extensive simulations, the proposed algorithms can reduce the average playback lag and cross-ISP traffic greatly.

IP address lookup is a challenging problem due to the increasing routing table size, and higher line rate. This paper investigates a new way to build an efficient IP lookup scheme using graphics processor units(GPU). Our contribution here is to design a basic architecture for high-performance IP lookup engine with GPU, and to develop efficient algorithms for routing prefix operations such as lookup, deletion, insertion, and modification. In particular, the IP lookup scheme can achieve O(1) time complexity. Our experimental results on real-world route traces show promising 6x gains in IP lookup throughput.

The benefit of network coding with respect to simplifying scheduling overhead for content distribution has been extensively studied in previous literature. However, the complexity of network coding increases as the content size scales up. In this paper, we study the tradeoff between scheduling overhead and coding overhead. To this end, we propose Swifter, a P2P content distribution scheme, which employs local-rarest-first segment scheduling and chunked network coding algorithms. In Swifter, content is divided into segments, which are further divided into blocks. Each peer schedules a local-rarest segment request from its neighbors. Network coding is then used for generating a reply block within the requested segment. Leveraging our real-world implementation and experiments, we find that Swifter has low coding overhead and can reduce average download time by up to 40% compared to existing work.

Nowadays, the real practice of network coding in wireline networks is focused on the P2P overlay networks. Although it can help to utilize network resources more efficiently, P2P network coding does not exhibit benefits in terms of the maximum throughput. Our work aims to implement network coding at the IP layer, which is an idea not fundamentally new, but with little real practice because of the enormous difficulties involved. In this paper we propose CODED IP, a protocol framework that plugs network coding into the current IP stack. Experiments on a 22-node testbed show that CODED IP provides multicast traffic with not only a significantly higher throughput than overlay network coding and naive IP multicast, but also a more balanced load distribution as compared with overlay network coding.