The Internet of Vehicles (IoV), a critical large-scale Internet of Things (IoT) application, faces a fundamental sustainability challenge stemming from the lifecycle mismatch between long-duration vehicular hardware and rapidly evolving software. This mismatch creates a widening “computational divide,” where aging vehicles with limited onboard resources cannot support modern data-intensive applications, thereby fragmenting the ecosystem and undermining its collective intelligence. To address this systemic issue, this paper proposes a novel lifecycle-aware task offloading framework. Instead of treating vehicular heterogeneity as a static liability, our framework transforms it into a dynamic, cooperative resource-sharing opportunity. At the core of this framework is a Deep Reinforcement Learning (DRL) agent deployed on resource-constrained vehicles, which orchestrates task offloading decisions to co-optimize for latency and energy consumption. A key innovation is the “Performance Capacity” metric, a multi-dimensional and predictive state assessment mechanism. This metric enables informed decision-making by intelligently fusing a vehicle’s static hardware profile, its dynamically predicted future workload, and its cooperation reputation. Furthermore, an integrated credit-based incentive mechanism is designed to ensure the long-term economic viability of this cooperative ecosystem. Simulation results demonstrate that our framework significantly reduces task latency and energy consumption, particularly under high-load conditions, offering a scalable and sustainable solution for the continuous evolution of heterogeneous IoV systems.

Anomaly detection in Industrial Internet of Things (IIoT) systems is crucial for ensuring operational safety and product quality. However, challenges such as heterogeneous sensor data, dynamic network structures, and complex spatio-temporal dependencies hinder existing methods. To address these issues, we propose a novel Transformer-based anomaly detection framework designed for dynamic IIoT networks. Our framework integrates self-supervised graph contrastive learning with multi-center Support Vector Data Description (SVDD) for the first time. Specifically, we construct temporal graph snapshots from normal time windows. Graph Attention Networks (GAT) are used to extract structural features, while a lightweight Transformer with temporal window attention captures long-range dependencies among sensor sequences, producing robust spatio-temporal node representations. To improve the model’s generalization without requiring labeled anomalies, a self-supervised contrastive loss is introduced. The resulting node embeddings are clustered via k-Medoids, and an SVDD sub-model is trained per cluster to define multiple hyperspheres representing normal behavior. During inference, the proximity of test representations to these hyper-spheres enables accurate and interpretable anomaly detection. Extensive experiments on real-world IIoT datasets, including SWaT and WADI, show our method outperforms state-of-the-art baselines in AUC and F1 score, demonstrating its effectiveness and practical value for IIoT security applications.

Large-scale Visual-Language Models (LVLMs) have achieved remarkable success in natural visual tasks, yet their application to industrial defect detection remains challenging due to two fundamental limitations: (i) the scarcity of large-scale industrial datasets that cover diverse defect categories across multiple domains, and (ii) the reliance on manual prompts (points, boxes, masks) that introduce subjective noise and lack text-visual interaction for fine-grained understanding. To address these challenges, we introduce a Large-Scale Multi-Modal Industrial Open-Closed benchmark (MMIOC-1 M) containing over one million samples across 14 super-categories, 29 industrial scenes, and 351 defect subcategories. To our knowledge, MMIOC-1 M is the first unified largest benchmark supporting both open-vocabulary and closed-set industrial detection, providing valuable pre-training data for LVLMs in industrial scenarios. Furthermore, we propose a Refined Text-Visual Prompt Network (RTVPNet) that incorporates three key innovations: (1) an expert-assisted domain projection mechanism that enables rapid adaptation of general vision models to industrial domains, (2) an energy-based sparse sampling strategy that automatically generates refined visual prompts without manual intervention, and (3) a bidirectional text-visual interaction module that enhances cross-modal semantic alignment and understanding. Extensive experiments demonstrate that RTVPNet achieves state-of-the-art performance on MMIOC-1 M, LVIS, and COCO benchmarks while maintaining computational efficiency.

Deep reinforcement learning (DRL) has shown significant potential in adaptive traffic signal control (ATSC) by adapting to real-time traffic conditions. However, controlling multiple intersections faces challenges, mainly due to the isolated actions of agents and non-stationary caused by other intersections. To address these issues, this paper proposes a novel knowledge collaboration-based actor-critic policy gradient (KCACPG) method to achieve cooperative traffic scheduling across multiple intersections. KCACPG includes a knowledge collaboration learning mechanism that allows heterogeneous agents to exchange knowledge across experience tuples, achieving globally optimal decision-making and coordination. KCACPG also integrates an off-policy prioritized experience replay mechanism to improve knowledge reuse efficiency and reduce the negative impact of knowledge transfer. Simulation results show that KCACPG converges quickly, generalizes to fluctuant traffic and load well, improves the network throughput by up to 17.8%, and reduces the pressure imbalance by up to 11.6% compared with the existing collaborative methods. The proposed method has significant implications for intelligent transportation systems and smart cities.

In this paper, we utilize GLM-4, a large language model, for decision-making in the realm of autonomous driving and compare its performance with the Deep Q-Network (DQN) and Graph Representation for Autonomous Driving (GRAD) method from reinforcement learning. Autonomous driving is a highly complex task, it needs the ability to make decisions in diverse environments. Thus, we aim to assess the potential of GLM-4 in various situations. GLM-4 has vast pre-trained knowledge that can have an inference based on contextual information. Our approach integrates the GLM-4 with the memory module that stores past experiences and leverages the pre-trained knowledge of GLM-4 to enhance decision-making under different driving scenarios. We tested the memory module with 5, 20, and 40 memory items, and conducted experiments using 1-shot and 3-shots experiences to analyze how accumulated experience influences the model's performance.

Consistency identification in task-oriented dialog (CI-ToD) typically consists of three sub-tasks: User Query Inconsistency (QI) identification, Dialogue History Inconsistency (HI) identification, and Knowledge Base Inconsistency (KBI) identification, which aim to determine inconsistent relationships between system response and user query, dialogue history, and knowledge base. Previous approaches focus on the exploration of deep learning models for CI-ToD. While these models achieve remarkable progress, they still rely on large amounts of labeled data, which is hard to achieve in real-world scenarios. Motivated by this, in the paper, we aim to explore large language models for CI-ToD, which do not require any training data. In addition, we further introduce a multi-agent collaboration framework (MAC-CIToD) to model the interaction across three sub-tasks in CI-ToD, including (1) Full Connection paradigm, (2) Cycle Connection paradigm, and (3) Central Connection paradigm, which effectively builds interaction across QI, HI, and KBI. Experiments on the standard benchmark reveal that our framework achieves superior performance. Additionally, we compare MAC-CIToD with the most advanced trained approaches and find that its zero-shot performance on most metrics even surpasses that of models after training on the CI-ToD dataset.

While traditional object detection methods have achieved significant success in recent years, their performance in detecting small objects in aerial remote sensing images remains unsatisfactory. Small objects often occupy only a few pixels, leading to a loss of fine-grained information. This paper proposes a dynamic feature and context enhancement network (DFCE) to address noise and pixel-level region weighting issues in small object detection. The DFCE network effectively detects small objects by dynamically selecting features within regions and establishing connections between local and global contextual information. The introduced dynamic multi-dimensional attention (DMA) module selects key information via a crossover mechanism and assigns different weights to highlight important features. Based on DMA, the regional feature processing (RFP) module and multi-dimensional pool transformer (MPT) module are developed to capture key information and contextual information, respectively. Experimental results demonstrate that the DFCE network improves average precision (AP) by 3.1%, 9%, and 2.2% on two remote sensing datasets and one conventional dataset, achieving an inference speed of 30 frames per second (FPS). Given these advancements, the DFCE model’s powerful key feature extraction and contextual association capabilities show strong potential for broader applications, including sign language recognition, defect detection in industrial settings, and more.

With the benefits of compact size and cost effectiveness, the solid-state LiDAR (SSL) has emerged as the preferred choice for the mass-produced vehicles with advanced driver assistance systems (ADAS). To ensure precise and dependable mapping for ADAS, it is essential to accurately fuse LiDAR and inertial measurements through precise extrinsic calibration. However, the existing extrinsic calibration methods primarily target the mechanical LiDAR and global navigation satellite system (GNSS)/inertial navigation system (INS), with limited research dedicated to the extrinsic calibration of SSL and GNSS/INS for the mass-produced vehicles. To achieve accurate calibration between the SSL and GNSS/INS, we develop a targetless extrinsic calibration method based on multiple constraints, called MCSGCalib. Specifically, we first propose a high-level geometric feature extraction approach based on the adaptive voxel to efficiently obtain the high-level geometric features for the SSL. Based on the extracted geometric features and ego-motion information, we construct multiple derived-friendly constraints, and fuse the motion constraints and geometric constraints into a unified manifold optimization framework in order to accurately estimate extrinsic parameters for the mass-produced vehicles, even in the presence of degenerate motions. To validate the performance of MCSGCalib, we conduct comprehensive experiments using the data collected from our mass-produced vehicles equipped with RS-LiDAR-M1, a recently released SSL extensively deployed in the mass-produced vehicles. The experimental results demonstrate the accuracy and robustness of the MCSGCalib in calibrating the extrinsic parameters between the SSL and GNSS/INS, showcasing its potential in enhancing the reliability and quality of the SSL and GNSS/INS fusion for ADAS.

Motivation:The prediction of drug-target interaction (DTI) from heterogeneous biological data is critical to predict drugs and therapeutic targets for known diseases such as tumor and bowel disease. The study of DTI based on drug representation learning can strengthen or integrate our knowledge of pharmacological and chemical phenomena. Therefore, there is a strong motivation to develop effective methods that can detect these potential drug-target interactions.Results:We have developed a novel Knowledge-Enhanced Multi-View framework (KEMV) to predict unknown DTIs from pharmacological data and chemical data on a large scale. The proposed method consists of two steps: (i) learning more comprehensive drug representations via the proposed multi-view attention mechanism, which bridges pharmacological and chemical information, and interactively summarizes the attention values depending on varying interactions between different pairs of drug features. (ii) predicting unknown drug-target interactions based on the drug and target representations. The method is tested on real-world dataset KEGG with three classes of important drug–target interactions involving enzymes, ion channels, and G-protein-coupled receptors. Our framework is proven to uncover potential DTIs with scientific evidences explaining the mechanism of the interactions through the processing of high-dimensional, heterogeneous, and sparse drug data.Availability:The originality of the proposed method lies in the attentive integration of pharmacological and chemical information for representation of drug candidate compounds and the prediction of drug-target interaction toward drug discovery in a unified framework. Our results are reproducible and and code is available at:https://github.com/YuanKQ/DTI-Prediction.

With the rapid development of deep learning, more researchers have attempted to apply nonlinear learning methods such as recurrent neural networks (RNNs) and attention mechanisms to capture the complex patterns hidden in stock market trends.Most existing approaches to this task employ an attention mechanism that primarily relies on the information extracted from input features but fails to consider the other important factors (e.g., trading volume and position), which can potentially enhance these attention-based approaches.Motivated by the observation, we extend the attention mechanism with features needed for stock performance prediction in this article.Specifically, we propose a volume-aware positional attentionbased recurrent neural network (VPA-RNN) for this task.First, we propose a generic method of adding position awareness to the attention mechanism.Next, the trading volume is incorporated into the original attention distribution to form a revised distribution.To evaluate the effectiveness of VPA-RNN, we collected real stock market data for stock indexes S&P 500 and DJIA, and the experimental results show that the proposed VPA-RNN can significantly outperform several existing highly competitive methods.

With the rapid development of deep learning, more researchers have attempted to apply nonlinear learning methods such as recurrent neural networks (RNNs) and attention mechanisms to capture the complex patterns hidden in stock market trends. Most existing approaches to this task employ an attention mechanism that primarily relies on the information extracted from input features but fails to consider the other important factors (e.g. trading volume and position), which can potentially enhance these attention-based approaches. Motivated by the observation, we extend the attention mechanism with features needed for stock performance prediction in this paper. Specifically, we propose a volume-aware positional attention-based recurrent neural network (VPA-RNN) for this task. First, we propose a generic method of adding position awareness to the attention mechanism. Next, the trading volume is incorporated into the original attention distribution to form a revised distribution. To evaluate the effectiveness of VPA-RNN, we collected real stock market data for stock indexes S&P 500 and DJIA, and the experimental results show that the proposed VPA-RNN can significantly outperform several existing highly competitive methods.

Rapidly growing healthcare big data are becoming a valuable resource for improving clinical activity through technology. Estimation of the severity of diseases and symptoms in electronic medical records is important because it allows physicians to more easily gain an understanding of medical documents. In this paper, motivated by the fact that these two tasks can benefit each other, we propose a clinical multi-task learning approach (CMTL) that integrates severity estimation and symptom name recognition into a unified framework. Specifically, CMTL consists of two key components: (i) a named-entity recognition model that learns rich knowledge-aware entity representations and classifies entity terms into predefined categories; (ii) a severity estimation model that provides the symptom severity stages of patients' self-reported symptom descriptions. These two tasks work on a shared text-encoding layer; then, a multi-head attention mechanism is proposed to learn the important information from different representation subspaces at different positions. Finally, we propose a shared label-transferring network to enhance their interaction in a multi-task learning setting. We created the first symptom severity estimation dataset (SymptomSE) and a gastrointestinal ontology to evaluate the effectiveness of our model. The experiments demonstrate the effectiveness of the proposed method and the improved performance of both tasks.

Password-based authentication is essential to any online service. It is normally powered by a database of user credentials, for example a RADIUS server. However, even with various indexing techniques (e.g., B+-tree), password-based authentication can still be resource-consuming on large-scale systems (e.g., Internet and IoT), and is thus vulnerable to distributed denial-of-service (DDoS) attacks.In this paper, we propose a cloud-based firewall that uses Bloom filters to pre-screen and reject suspicious requests with wrong password before they reach the authentication server. The main challenge is the security of the firewall because it can be operated by a third party, so the Bloom filters might be accessed by adversaries to assist their brute-force password guessing.To ensure security, we start with the assumption of trusted cloud server and design a key-based semantic secure Bloom filter (KSSBF) for the best efficiency. We then design a generically secure Bloom filter (GSBF) for non-trusted cloud servers, which is key-independent and with strictly provable security. Through theoretical and empirical analysis, we show both of them can mitigate malicious requests without compromising the security of passwords.

Network representation learning (NRL), which aims to map nodes in a network into low-dimensional vectors, has attracted wide attention due to its potential on various net-work applications. Recently, attributed network embedding that incorporates both network structure and node attributes has shown a good performance in NRL. However, most existing methods cannot achieve promising results in sparse attributed networks because of the lack of structural information. In order to further alleviate the negative effect of sparseness, we propose a novel model named AANI (Adaptively Aggregating Neighborhood Information). Specifically, AANI exploits the information of current node and its neighborhood in a smoothing way to obtain more informative node representations. At the same time, AANI introduces attention mechanism to adaptively aggregate the smoothed information according to their respective importance. We conduct extensive experiments on three real-world datasets which demonstrate that AANI outperforms the state-of-the-art embedding methods in sparse attributed networks.

With the emerging of NVM (Non-Volatile Memories) technologies, NVMM-based (Non-Volatile Main Memories) file systems have attracted more and more attention. Compared to traditional file systems, most NVMM-based file systems bypass the page cache and the I/O software stack. With the new mmap interface known as the DAX-mmap interface (DAX: direct access), the CPU can access the NVMM much faster by loading from/storing to it directly. However, the existing file system benchmark tools are designed for traditional file systems and do not support the new features of NVMM-based file systems, so the returned results are very often not accurate. In this paper, a new benchmark tool called NVMFS-IOzone is proposed. The behavior of the tool is redesigned to reflect the new features of NVMM-based file systems. The NVM-lib from Intel is used instead of traditional msync() to keep data consistent when evaluating the performance of the DAX-mmap interface. Experimental results show that the new benchmark tool can reveal a hidden improvement of 1.4~2.1 times in NVMM-based file systems, which cannot be seen by the traditional evaluation tools. The data paths of direct load/store to NVMM and bypassing CPU cache are also provided to support the new ##features of NVMM-based file systems for multidimensional evaluation. Furthermore, embedded cleaning-ups has also been added to NVMFS-IOzone to support convenient evaluation consistency, which benefits both NVMM-based and non-NVMM-based file system benchmarking even for quick and casual tests. The whole experimental evaluation is based on real physical NVMs rather than simulated NVMs, and the experimental results confirm the effectiveness of our design.

Software switches are being deployed in SDN to enable a wide spectrum of non-traditional applications. The popular Open vSwitch uses a variant of Tuple Space Search (TSS) for packet classifications. Although it has good performance on rule updates, it is less efficient than decision trees on lookups. In this paper, we propose a two-stage framework consisting of heterogeneous algorithms to adaptively exploit different characteristics of the rule sets at different scales. In the first stage, partial decision trees are constructed from several rule subsets grouped with respect to their small fields. This grouping eliminates rule replications at large scales, thereby enabling very efficient pre-cuttings. The second stage handles packet classification at small scales for non-leaf terminal nodes, where rule replications within each subspace may lead to inefficient cuttings. A salient fact is that small space means long address prefixes or less nesting levels of ranges, both indicating a very limited tuple space. To exploit this favorable property, we employ a TSS-based algorithm for these subsets following tree constructions. Experimental results show that our work has comparable update performance to TSS in Open vSwitch, while achieving almost an order-of-magnitude improvement on classification performance over TSS.

Per-flow frequency estimation plays a fundamental role in network measurement. As a probabilistic data structure, sketch has been extensively investigated and used for per-flow frequency estimation, but most sketch-based proposals in previous literatures cannot achieve high accuracy and high speed simultaneously. Moreover, because each insertion to a sketch causes increment in multiple entries, the over-estimation error will accumulate quickly over time. In this paper, we propose Cuckoo Counter, a compact and accurate framework for per-flow frequency estimation, which employs three novel ideas: (1)kicking out conflicting flows instead of using multiple entries counts to improve accuracy; (2)using different sizes of entries to insulate mice flows from elephant flows, which can handle the skewed data streams efficiently and improve memory utilization; (3) a Cuckoo-like replacement strategy for mice flows, so as to maintain accurate records for elephant flows. To verify the effectiveness and efficiency of our framework, we compared it with two well-known sketches as well as the recent proposed Augmented sketch and Pyramid sketch. Extensive experimental results on three different types of test datasets show that Cuckoo Counter outperforms these sketches considerably.

For machine-learning-based network traffic classification, we usually need large number of correctly labeled samples (ground truth) for model-training to get high accuracy. However in practical environments obtaining ground truth may be time-consuming and needs massive manual work, so achieving higher accuracy at low labeling rate is still a tricky problem. In this paper we propose a novel Graph Convolutional Network (GCN) based network traffic classification method named GCN-TC. We combine the traffic trace graph with statistical features in GCN model-training, so as to take advantage of both of them to achieve higher classification accuracy with very few labeled data. Experiment results show that the proposed model achieves the best performance with 13% higher in Fl-score and 7.7% higher in accuracy than the best method of the rest.

Cuckoo hashing is widely used for its worst-case constant lookup performance even at very high load. However at high load, collision resolution will involve lots of probes on item relocation and may still fail in the end. To address the problem, we propose an efficient Cuckoo hashing scheme called Multi-copy Cuckoo or McCuckoo. Different from the blind kick-outs of standard Cuckoo hashing during a collision, we can foresee which way to successfully kick items by using multiple copies. Furthermore, with the knowledge of how many copies each item has in the table, we can identify impossible buckets and skip them during a lookup. In order to avoid expensive rehashing during insertion failures, McCuckoo also supports more efficient stash strategy that minimizes stash checking. McCuckoo uses simple logic and simple data structure, so it is suitable for both software and hardware implementation on platforms where intensive access to the slow and bandwidth limited off-chip external memory is the main bottleneck.

Multi‐field packet classification is not only an indispensable and challenging functionality of existing network devices, but it also appears as flow tables lying at the heart of the forwarding plane of software defined networking age. Despite almost two decades of research, algorithmic solutions still fall short of meeting the line‐speed of high‐performance network devices. Although decomposition‐based approaches, such as cross‐producting and recursive flow classification (RFC), can achieve high lookup rate by performing a parallel search on chunks of the packet header, both of them suffer from memory explosion problem during aggregation. In this study, the authors propose an HybridRFC, a memory‐efficient recursive scheme for multi‐field packet classification. By addressing the embedded problem of the RFC caused by uncontrollably expanded cross‐product tables, HybridRFC can not only reduce the memory consumption to a practical level but also improve pre‐processing performance significantly. Experimental results show that the memory requirement of HybridRFC is two orders of magnitude less than RFC, as well as three orders of speed‐up on the performance of table building on average.

This paper describes the design of a new software zero-copy network framework for NVM-assisted key-value stores, which directly stores and persists transactions from network into raw non-volatile memory used as write-ahead cache for data consistency. NVM is fast and bit-addressable which makes it the perfect choice for transient transaction log persistency than hard disks or even Flash drives, but its limited write cycle requires wear-leveling during direct access. However, popular RDMA-based zero-copy transmission normally needs to have the remote memory address beforehand and cannot cope with the address changing caused by wear-leveling easily. The software zero-copy solution proposed in this paper is designed with the awareness of NVM wear-leveling and log metadata management. Simulation results show that the new network framework improves performance by over 200× in throughput and decreases latency by more than 20× comparing to the traditional socket and hard disk based solution. When both equipped with NVM, the zero-copy network stack improves performance by 18 to 62% in throughput and 40 to 81% in latency comparing to the standard socket and with the lowest CPU consumption.

Software-Defined Networking (SDN), which separates the control plane and data plane, is a promising new network architecture for the Future Internet. OpenFlow is the de facto standard which defines the communication protocol between the controller and switches. The most challenging issue in OpenFlow switches is the lookup of multiple OpenFlow tables. The lookup of OpenFlow tables is so complicated that the state-of-the-art research are still focusing on the design of lookup pipeline architecture, and there is no specific algorithm for the lookup of OpenFlow tables. In this paper, we revise the long-pipeline architecture of OpenFlow 1.4 to a 5-stage pipeline architecture to make a trade-off between flexibility and implementability, and decompose the lookup of OpenFlow tables into three kinds of lookup: longest prefix matching (IP lookup), multi-field matching (packet classification), and exact matching. Then we design new algorithms for packet classification, because the state-of-the-art solutions for them seldom support fast update which is highly demanding for OpenFlow. The other two kinds of lookups can be well handled by state-of-the-art. Experimental results show that our proposed algorithms work excellently, and outperform state-of-the-art solutions.

Packet classification is one of the core functions in present-day networking applications. Various classification algorithms have been developed, including tuple space search (TSS) which is used in the Open vSwitch due to its flexibility and scalability. In order to solve the problem of excessive memory access to the vast rule-set stored off-chip, many algorithms adopt bloom filter (BF) to reduce unnecessary accesses. However, deteriorated false positive and increased collisions at high load ratio may still be a problem. Here in this paper a new tuple pruning based packet classification algorithm is proposed that can achieve much reduced off-chip access per-lookup even at high load by adopting a better compact structure on-chip and Cuckoo hash off-chip. At the same time, we can accelerate the classification rate further by optimizing the search sequence of tuples making it more likely to find the matching rules earlier. Experimental results show that with the proposed mechanism the off-chip table lookup time is only 1/3 to 1/4 of the baseline BF- based scheme.

The emerging memory technologies, such as phase change memory (PCM), provide chances for highperformance storage of I/O-intensive applications. However, traditional software stack and hardware architecture need to be optimized to enhance I/O efficiency. In addition, narrowing the distance between computation and storage reduces the number of I/O requests and has become a popular research direction. This paper presents a novel PCMbased storage system. It consists of the in-storage processing enabled file system (ISPFS) and the configurable parallel computation fabric in storage, which is called an in-storage processing (ISP) engine. On one hand, ISPFS takes full advantage of non-volatile memory (NVM)’s characteristics, and reduces software overhead and data copies to provide low-latency high-performance random access. On the other hand, ISPFS passes ISP instructions through a command file and invokes the ISP engine to deal with I/O-intensive tasks. Extensive experiments are performed on the prototype system. The results indicate that ISPFS achieves 2 to 10 times throughput compared to EXT4. Our ISP solution also reduces the number of I/O requests by 97% and is 19 times more efficient than software implementation for I/O-intensive applications.

Disk-based large scale graph computation on a single machine has been attracting much attention, with GraphChi as one of the most well-accepted solutions. However, we find out that the performance of GraphChi becomes I/O-constrained when memory is moderately abundant, and from some point adding more memory does not help with the performance any more. In this work, a greedy caching algorithm GSC is proposed for GraphChi to make better use of the memory. It alleviates the I/O constraint by caching and delaying the write-backs of GraphChi shards that have already been loaded into the memory. Experimental results show that by minimizing unnecessary I/Os, GSC can be up to 4x faster during computation than standard GraphChi under memory constraint, and achieve about 3x performance gain when sufficient memory is available.

Cost computation, cost aggregation, disparity optimization, and disparity refinement are the four main steps for stereo matching. While the first three steps have been widely investigated, few efforts have been taken on disparity refinement. In this paper, we propose a color image-guided disparity refinement method to further remove the boundary-inconsistent regions on disparity map. First, the origins of boundary-inconsistent regions are analyzed. Then, these regions are detected with the proposed hybrid-superpixel-based strategy. Finally, the detected boundary-inconsistent regions are refined by a modified weighted median filtering method. Experimental results on various stereo matching conditions validate the effectiveness of the proposed method. Furthermore, depth maps obtained by active depth acquisition devices like Kinect can also be well refined with our proposed method.

Hash table is used in many areas of networking such as route lookup, packet classification, per-flow state management and network monitoring for its constant access time latency at moderate loads. However, collisions may become frequent at high loads in traditional hash tables, which may lead the access time complexity to be linear and intolerable to applications like high-speed route lookups. While some schemes were proposed to help resolve this problem and most of them may achieve O(1) average memory access per lookup, very few of them are able to cut down the access to a deterministic single one. In this paper, we design a structure called deterministic and efficient hash table (DEHT). In DEHT, a novel data structure on on-chip memory is built, with the help of which the off-chip memory access can be decreased to a single one at most per lookup even when the load of the hash table is very high. What's more, the on-chip data structure also plays a similar role as Bloom Filter to do membership screening, which can avoid most lookups of nonexistent items of the hash table visiting the off-chip memory. Through theoretical analysis and simulations, we show that our scheme is faster than other schemes in lookup operations; the usable load of the off-chip hash table, the memory efficiency and the false positive rate of the on-chip data structure are also favorable.

Data center networks are high-bandwidth, low-latency networks with clear-defined topology and busy internal data exchange traffic. TCP is widely used in data centers to transport data, however its default behavior during the flow initialization phase may be unsuitable for data center networks and affect the responsiveness of data transmission, especially when the flow is short. Considering that data center flows happen frequently but only between limited number of data center servers connected by a rather stable physical topology, in this paper we propose to use the RTT records from previous transmissions to help TCP calculate more suitable initialization parameters when a new flow starts, so as to avoid unnecessary lengthy timeouts that has heavy consequences on the overall performance, especially on short flows. Experiments show that our algorithms can indeed eliminate the long timeouts caused by early packet losses, and improve the responsiveness and stability of data transmission in a data center networking environment.

With the maturity of the cloud infrastructure, many data are moved to the cloud. However, availability and security of the cloud data still remain the major concerns. To relieve these concerns, it has been proposed to disperse encoded data redundantly across multiple independent cloud providers, so that at least a certain number of data fragments are required for data recovery, thus called cloud-of-clouds (CoC) systems. As for system performance, there is always a trade-off among sufficient security or high availability with different settings on the number of fragments and the threshold value. Here in this paper we extend the scenario and propose a comprehensive hybrid model that provides a holistic framework taking into account the diverse reliability and security characteristics of the storage providers in a CoC system to enhance the enforcement of data protection. The availability and security of the system in our model is analyzed theoretically in a systematic way. Simulation results also show that with our model, by exploiting the asymmetrical characters of the CoC system, both security and availability can be achieved satisfactorily.

TCP Incast is one of the typical datacenter problems that may affect network efficiency in a many to one transmission session, in which the completion time of the whole session depends on the last finishing flow. In this paper a session-aware mechanism is proposed that intelligently chooses only the leading flows to slow down at the presence of congestion, so the lagging ones will have a higher chance to catch up to achieve better session completion time and goodput. Compared to existing solutions, the proposed one introduces no change to the TCP protocol and no extra messaging for the internal status of either TCP or the intermediate switches. Experimental results show that the proposed mechanism can indeed balance the progress among Incast flows and achieve a shorter session time and higher goodput.

In Named Data Networking (NDN), the speed of name lookup operation and the scalability concern impact deeply on the performance of the forwarding system. NDN names can contain unbounded number of components which may result in intolerable lookup time. And since routers cannot have unlimited table size to keep up with the unbounded nature of application data namespace of NDN, it may lead to missing records for valid data names in the routing table. In this paper, we propose a port information assisted trie or P-trie to help solve these two issues: name lookup process is accelerated with the help of the port information on nodes of the trie, and packets with names not in the routing tables are forwarded to a most promising port based on the analysis of existing entries. We implemented P-trie in a multi-aligned transition array (MATA) to reduce the memory footprint. Experiments show that the name lookup rate of our P-trie can be times faster than other trie based methods.

Due to the explosive growth of multimedia content (especially videos) over the Internet, content-centric networking (CCN) is proposed to remit the problems of modern bandwidth-intensive Internet usage patterns. Additionally, the current streaming media coding is designed for video service for IP networks. However, there are few researches who concentrate on efficient streaming media coding for the CCN pattern. It motivates us to find a suitable video coding to improve the performance of CCN. This paper proposes an advanced caching management strategy by combining the Multiple Description Coding (MDC) to the content items, termed as MDC-Ca. The core parts of CCN, e.g., location-independent naming, name-based routing and in-network caching strategy, are all adjusted to the content communication. Moreover, MDC-Ca can deal with the ruleless content chunk distribution in the caching of network nodes. MDC-Ca was studied in a mathematical analysis model and the scheme performs well in the simulation. Further, we perform the emulation in a practical network. The experimental results from both simulations and practical emulations show the superiority of the proposed caching management strategy.

Identity centric networks (IdCN), in which packets are destined to user identities, is one of the future network architectures under discussion. Since user mobility is already widely observed in the current Internet and seen as one of the main features of the future Internet, seamless mobility needs to be kept in mind for whatever routing mechanisms developed for IdCN. However, seamless handoff techniques from the IP world won't work because no Care-of-Address or the alike is available in IdCN to help quickly locating the mobile user's new position, therefore it is difficult to redirect the packets in transit before the routing protocol re-finds the moved users. In this paper a dynamic ring mechanism that does not rely on but can work with any IdCN routing protocol is proposed to support seamless handoff in IdCN. During a handoff, the mobile user makes use of this ring to keep its global connectivity, and leaves after the handoff finishes. Appropriate timing of hooking to and leaving the ring is crucial to keep the overhead at minimum. Experiments show that the proposed mechanism can indeed prevent most packet drops and keep the flow running during handoffs at reasonably lower cost than simple retransmission.

Joining objects from different web sources and returning top-k combinations is a research topic with much attention. Such techniques could be used in location based scenarios, which for one example, help plan a wonderful night by finding a good combination of hotel, restaurant and theater. Challenge to such techniques is that in a good combination, each individual object should be good enough and, of equal or even greater importance, subject to some given criteria, such as within a given range and close to each other. Proximity rank join method is one way to settle this problem. It takes advantage of sorted access of inputs, does not rely on specialize data structures to determine spatial closeness, and has efficient pull and bound strategies to avoid reading too much inputs before finding the top-k answers. In this paper, we propose a detailed scheme using the proximity rank join model that is optimized for location based purposes. It will be shown that our scheme indeed reflects user's desire better and outperform the naïve Euclidean distance scheme.

In a content centric network, packet forwarding is performed over data names instead of IP addresses. Since data names are order-of-magnitude larger in number and complexity, CAM-based or Trie-based techniques are not applicable any more. and new forwarding schemes are proposed to solve the problem. These schemes use hashing to store the large routing table for named data into relatively abundant off-chip memory, and use some on-chip Bloom filter to minimize expensive off-chip memory access by quickly screen out table lookup queries for unrecorded names. In this paper we propose to add a `summary vector' to the on-chip Bloom filter that can help in constructing an efficient off-chip hash table for better storage and lookup performance: a dynamic collision-free hash table that only needs to read into only one routing record for any lookup queries.

Over the past decade and more, several accurate but complex traffic models have been developed by just as many researchers. Among the most accurate models are those based on mathematical principles that are able to model the multifractal nature of network traffic. Unfortunately these models are hardly even usable by network engineers because they lack a connection to operational parameters that can easily be estimated based on the knowledge of the network. In this paper, we try to bridge the gap between high-level parameters describing the concerning network and the inputs the modern traffic models need to generate artificial traffic. We first build a model of the behavior of TCP over the duration of a flow, and we then approximate this behavior based on simple parameters such as for example the packet transmission time and the RTT.

A large number of research articles are devoted to queuing theory and queuing systems. Most of these articles employ a continuous representation of network traffic, in the form of timestamps or interarrival times. In this, there is a contradiction with more recent traffic models capable of capturing the multi-fractal nature of network traffic e.g. the conservative cascade model. These models often represent packet streams in a discrete way by calculating the bin count vector. Directly describing the effect of queuing systems on the variance-time behavior of this discrete representation of traffic is relatively unexplored terrain. This paper presents and analyzes some qualitative results on the altering of a bin count vector when passing it through a queuing system. After the detailed analysis of a basic fixed service time queue, some considerations are made on real networking components. This leads to extending the basic queue model with variable service times, based on the packet size distribution. Finally, the influence of the shape of this distribution on the queuing effects is studied.

A large number of research articles are dedicated to queuing theory and queuing systems. Most of these articles employ a continuous representation of network traffic, in the form of timestamps or interarrival times. In this, there is a contradiction with more recent traffic models capable of capturing the multi-fractal nature of network traffic e.g. the Conservative Cascade model. These models often represent packet streams in a discrete way by calculating the bin count vector. Directly describing the effect of queuing systems on the variance-time behavior of this discrete representation of traffic is relatively unexplored terrain. This paper presents and analyzes some qualitative results on the altering of a bin count vector when passing it through a basic queuing system. The scenario is also extended with the influence of correlated background traffic.

TCP-friendly rate control(TFRC) is an equation-based congestion control mechanism that competes fairly with TCP but has a much lower throughput variation, which makes it a better choice for streaming over the Internet. It is known that in a mobile network environment, after a handover TFRC can overshoot or under utilize the new link if the conditions there differ from those of the old link. There are different factors that affect the TFRC performance during a handover. In this paper we focus on the impact of the change in the round trip time, an aspect that is largely overlooked in the literature compared with bandwidth disparity. We find that even if the same bandwidth is assured on the new link, changes in the link latency will still cause TFRC performance degradation after the handover. A fast link adaptation mechanism is thus proposed to address the problem. Simulations show that this mechanism helps TFRC to adjust quickly to the new link without consecutive packet loss or long-time bandwidth under utilization.