ABSTRACT. The second wave of cloud computing, named network cloudification, in the forms of SDN (Software Defined Networking), NFV (Network Function Virtualization), and 5G-MEC (Mobile Edge Computing), is to centralize and virtualize networking into data centers. It enables operators to offer NaaS (Networking as a Service) with much lower CAPEX and OPEX with larger flexibility because devices become simpler, the number of administrators is less, and service orchestration is easier. It turns parts of communications currently done in hardware into computing done in software. However, the host of these data centers would not be Google-like super data centers as they are too far away from subscribers. The latency requirement of 10ms and 1ms decentralizes cloud computing down to edge and fog computing with CORD (central offices re-architected as data centers) and cellular base stations for SDN-NFV and 5G-MEC, respectively. In this talk, we first argue why, where and when SDN, NFV, 5G-MEC would prevail, and then illustrate how to make it happen with OpenFlow, SC (Service Chaining), NSH (Network Service header), etc. Then we examine how latency requirement dominates this virtualization game by listing key questions to answer in resource allocation in the architectures of SDN, NFV, and 5G-MEC. Their answers are mostly unknown now but would benefit the architects and developers of OpenFlow switches, SDN controllers, SDN-NFV apps, NFV data centers, MEC-enabled base stations, and operator’s infrastructure in general.

ABSTRACT. High-speed metro and access networks are indispensable to provide broadband services, especially bandwidth-hungry services and future 5G mobile networks to end users. Advanced wireline and wireless network solutions with ultra-high capacity, low transmission delay, high energy efficiency, and low cost should be developed to support a wide range of applications and requirements in future mobile networks. A convergence of wireline and wireless networks should also be considered to simplify the network configuration, management, and to realize new features and functions. Advanced photonic-based transmission systems are thus of critical importance to facilitate future mobile and wireless communications. In addition, high-performance imaging would play an important role to enhance safety and security of public infrastructure including airports, railways, etc. High-resolution radars can be constructed using frequency in the millimeter-wave and terahertz-wave bands where wide radio frequency resource is available. To cover a wide coverage range, a combination of radio and optical technologies via radio-over-fiber technology would be very useful to distribute and collect radio-wave information. This tutorial will present technical challenges and enabling system/device technologies for future mobile transport networks, including mobile backhaul and fronthaul, and photonic-based millimeter-wave and terahertz-wave radar systems and related technologies for applications to public services, disaster recovery, etc.

ABSTRACT. In this study, we research issues emerging in developing a cost-effective smart farming system. By proposing a system architecture and relevant solutions, we successfully integrate different modules related to sensing systems, communication systems, and data analytics into a whole system that not only monitors the farm environment but also performs remote automation and user control. Moreover, we make the smart farm system smarter with a knowledge base attached. Our smooth-running prototype for the oyster mushroom and hybrid maize culture verifies the unified integration and demonstrates that farmers are no longer worried about how to manage and control their farms at distance while enjoying their profits. Furthermore, our Smart Farm IoT Kit is designed with cheap but effective devices and lightweight communication. Last but not least, we further discuss other challenges inevitable for commercialization study.

ABSTRACT. Today, our modern living world is covered by ubiquitous systems that offer the ability to assess, understand and handle the environmental indicators of our urban environment as well as our delicate ecology and natural resources. The sudden increase of such systems creates the notion of Internet of Things (IoT). However, the variety of components and wide-area deployments in IoT create a disadvantage point: the heterogeneity problem. Such system has multiple heterogeneous wireless communication solutions with multiple access technologies such as bluetooth, wifi, zigbee, cellular, MANET, etc. Concretely, the effectiveness is revealed by a variety of access technologies that are working on a common core network with a common policy for every type of access network. The challenge is how to manage this heterogeneous network in a dynamic context with an open and distributed infrastructure. One of the most efficiency solutions for this issue is the Software-Defined Network approach (SDN). This paper proposed a new SDN-based architecture with a centralized controller that has a capacity of self-observing and adapting. The SDN controller has the ability to incorporate and support user requests to classify flow scheduling over task-level. Besides, the paper creates an Ontology for analyzing user's request and based on the Lagrange relaxation theory for a heuristic routing algorithm. The experimental works showed that the proposed solution yielded impressive and good results.

ABSTRACT. Urban flood management and mitigation is one of the critical missions to maintain a stable city development. Most of the current available flood mitigation approaches are using data from fixed-site sensor systems which are not only costly but also limited in their coverage. Crowdsourcing is one of the potential approaches for addressing the coverage issue where flood data is collected and shared by crowds available everywhere. However, existing crowdsourcing-based approaches to flood mitigation systems lack thorough solutions on data validity. This paper proposes an appropriate crowdsourced data validity scheme which relies on both the scores evaluated by other users in the crowds and the accumulated reputation obtained by the user who share the data. The paper also presents a practical mechanism to cluster reported data based on spatial and temporal information to improve the effectiveness of the data analytics. The proposed approaches have been implemented in an urban flood mitigation platform prototype running on both the IOS and the Android mobile devices. This prototype preliminarily shows the appropriateness and the usefulness of the proposed solutions.

ABSTRACT. In this paper, we propose a control system and algorithm that can send and receive data between smartphone and IoT device without using a separate connection using WiFi and Bluetooth beacon signals. The existing IoT control technology can be controlled by directly connecting to the IoT device through a WiFi and Bluetooth network in a smart phone, or by connecting to a gateway or an AP connected in common. Existing technologies have incurred overhead such as current consumption and network occupation required for connection and connection maintenance for services and products that do not require continuous information reception or control. In order to reduce the overhead of existing technology, this paper proposes a method of transmitting control information by omitting a separate connection process by transmitting a beacon signal to an IoT device when a user inputs a control command. Soft AP and low power Bluetooth (BLE) are used in smartphone of Android platform and evaluated in Home IoT Network environment. As a result of implementing the proposed algorithm and measuring the performance, there is no overhead such as current consumption and network occupation required for connection and connection maintenance occurring in existing technology. The system using Soft AP was controlled at speed equivalent to that of existing technology within 8 meters in terms of control speed and improved about 10% at a distance of 8 meters or more. The system using BLE showed about 30% improvement in overall distance.

ABSTRACT. In multievent wireless sensor networks (WSN) like smart kindergarten, forest fire alarm system, vehicle tracking system, many events have different QoS (Quality of Service) requirements such as latency, rate, reliability, etc. Most of researches in this area have just dealt with one or two QoS requirements or one QoS requirement with several priority levels, there has not been any research supported multi QoS requirements for multievent WSN. This paper proposes a novel solution to meet the new and diverse requirements for multievent WSN called CSM. By combining hybrid routing protocol and packet delivering scheme, our proposed solution enables multievent WSN support multiple QoS requirements such as latency and reliability. Simulation results show that, with multiple QoS requirement events simultaneously occurring in the network, the proposed scheme brings better network performance in terms of latency (reduces 20% delay) for lower latency requirement events and lower packet error rate for higher reliability requirement events than other coexisting events.

ABSTRACT. Since 2000, when clustering with side information is introduced in the first time, so many semi-supervised clustering algorithms have been presented. Semi-supervised clustering, that integrates side information (seeds or constraints) in the clustering process, has been known as a good strategy to boost clustering results. In general, semi-supervised clustering focuses on two kind of side information including seeds and constraints, not much attention was given to the topic of using both seeds and constraints in the same algorithm. To address this problem, in this paper, we extend the semi-supervised graph based clustering (SSGC) by embedding both constraints and seeds in the clustering process; our new algorithm is called MCSSGC. Experiments conducted on real data sets from UCI show that our method can produce good clustering results compared with SSGC.

ABSTRACT. Finding similar two-dimensional shapes is an important problem with application in various domains. Shape motifs are approximately repeated shapes within image collections. Two-dimensional shapes can be converted to time series and the problem of discovering motifs in shape time series can be solved by exploiting some appropriate techniques from time series motif discovery. However, so far there exists one method proposed by Xi et al. which uses Euclidean distance to deal with this problem. As for shape data, Dynamic Time Warping (DTW) distance is more suitable and brings out more accurate results than Euclidean distance in similarity matching. Nevertheless, there are two difficulties with DTW distance in image matching: DTW distance incurs high computational complexity and image matching requires that the distance measure must be invariant to rotation. In this work, we propose a fast method for discovering shape motifs in a database of shapes using rotation invariant DTW distance. Our method is not based on Random Projection algorithm but employs a new approach for motif-discovery: clustering-based. Experimental results showed that our proposed method for shape motif discovery under DTW performs very efficiently on large datasets of shapes while brings out high accuracy.

ABSTRACT. Regular path queries (RPQs) are widely used on a graph whose answer is a set of tuples of nodes connected by paths corresponding to a given regular expression. Traditional approaches for evaluating RPQs are restricted in the explosion of graph size and/or highly complex query (e.g., nested query). Consequently, evaluating an RPQ on a large graph often takes high cost, causing substantial memory spaces and long response time. Recently, cost-based optimizations of RPQs have been proved to be effective when they are applied to large graphs. However, these techniques could not guarantee the minimum evaluation cost all the time. Therefore, estimating the evaluation cost of RPQs is an important topic which opens the way to cost-based graph query processing such as query optimization and parallel query evaluation. In this paper, we present a novel approach for estimating the evaluation cost of RPQs on large graphs. Our method exploits graph schema to make a so-called USCM (Unit-Subquery Cost Matrix), which presents the evaluation cost of the unit-subqueries (i.e. every smallest possible subquery). We propose some cost functions based on USCM to estimate the evaluation cost of an RPQ by decomposing it into a set of unit-subqueries. We also present a case study which applies our proposed idea for parallel RPQs evaluation. Experimental results show that our estimation method obtains high accuracy approximately 87% in average. Moreover, two comparisons with automata-based and rare label based approaches demonstrate that USCM-based approach outperforms traditional ones.

ABSTRACT. To date, entity coreference resolution and sentiment analysis still exist as independent problems, but they are both popular research topics in the community of natural language processing. However, the combination of those two problems, even though very potentially useful, has not been getting much attention. This paper addresses this issue by proposing an approach applying knowledge base to resolve coreference between two kinds of entities, objects and aspects, which are embedded with sentiment information. The knowledge base is introduced as a sentiment ontology. In addition, the paper also proposes a model of ontology-based coreference resolution in sentiment analysis for English text. Finally, we also discuss evaluation methods applied for our model and the results obtained.

ABSTRACT. In this paper, a solution of storing and inferring on citation network using graph database and topic model is proposed. Citation network is a very large directed graph containing nodes and edges. Each node is a paper and each directed edge is a link between paper and its citing papers. In citation network analysis, each node usually contains basic properties of paper such as paper ID, publication year, paper title, authors. In this research, we propose one more property called “topics”. This property contains topic distribution of a specific paper which is gained by Gibbs sampling of LDA model. After that, we propose a new approach to store and manage the citation network using graph database. Finally, we use the graph query language to develop some functions of citation network analysis and visualize topic propagation through the network. We also compare our approach with the traditional method in which the relational database is used to store and manage citation network. By experiment, the performance of our approach is higher than the traditional one.

ABSTRACT. This paper introduces an adaptive large neighborhood search (ALNS) meta-heuristic to solve the multi-trip multi-traffic pickup and delivery problem with time windows and synchronization (MTT-PDTWS). With adaptive destroy and repair operators, it learns over time which destroy and repair neighborhoods are the most effective and governs the neighborhood selection biased toward highly effective one. The computational experiments display the impacts of these operators on the solution quality and the performance of the ALNS in comparison to the only existing tabu search methodology addressing the MTT-PDTWS.

ABSTRACT. Nowadays, database security problem is increasingly more interested in the data mining field. How to can exploit legitimate data and avoid disclosing sensitive information. Tere have been many approaches but outstanding among them is privacy preservation in association rule mining to hide sensitive rules. In recent years, a meta-heuristic algorithm is quite effective for this purpose, which is the cuckoo optimization algorithm (COA4ARH). In this paper, an improved proposal of the COA4ARH to minimize the side effect of missing non-sensitive rules is introduced. The main contribution in this study, we propose a new pre-process stage to determine the fewest number of necessary transactions for the process of initializing an initial habitat, thus the restriction of modified operation on the original data. To evaluate the effectiveness of proposed method, it was conducted some experiments on real datasets. The experimental results show that improved approach has higher performance compared to the original algorithm.

ABSTRACT. This paper studies Meta-heuristic approaches to solve a nurse districting problem for a public medical clinic in Canada. Briefly, the territory covered by the clinic is divided into districts, each district is associated with some nurses, and each nurse is assigned to one district. Because the list of patients is modified regularly which leads to the change of patient-nurse assignment, it becomes difficult to maintain the workloads equilibrium among nurses. The objective consists in maintaining similar workloads among nurses while new patients are being added constantly. The continuity of care is obtained by keeping the same follow-up nurse of each patient as much as possible and avoiding moving too many patients from one nurse to another. We propose a mathematical model to formalize the districting problem and a Tabu search algorithm to solve the problem. Numerical results are provided based on real data and randomly generated data, with simulations to mimic the practical context.

ABSTRACT. In this paper, we propose a partial search approach for subword-based keyword search (KWS) systems. The proposed approach addresses the problem of high miss rate in the conventional full sequence matching approach by retaining detections that only contain some partial sequences. To control the increase of false alarm (FA), we propose two pruning methods, global threshold pruning (GTP) and keyword-specific pruning (KSP). The former uses the same threshold for all keywords; while the latter gradually adjusts a specific threshold for each keyword. Experimental results on the NIST OpenKWS15 and OpenKWS16 Evaluations show that partial search using KSP is effective. For out-of-vocabulary keywords, our proposed approach reduces the miss rate up to 21.0\% absolute and improves the Actual Term Weighted Value (ATWV) up to 19.2\% absolute over the full search baseline

ABSTRACT. Many real-world applications can be expressed as constraint satisfaction problems (CSPs), although hardly any problems are originally given by SAT formulas. Nevertheless, to benefit from powerful SAT solvers, many SAT encodings of CSPs have been studied recently. Such encodings should not only be effectively generated, but should also be efficiently processed by SAT solvers. In this survey we present a novel approach on how to encode a finite-CSP domains into a SAT instance in a comprehensive and precise way. The paper also provides a informative comparison among SAT encodings in term of the number of variables and clauses required, and the strength of consistency achieved by unit propagation.

ABSTRACT. As people say "Data is the new oil," Big data is expected to make a large impact on our society and economics by mining hidden knowledge and rules from the data. In particular, the structure of the real world data is changing from traditional relational data model to more generalized graph data model, as the web and social media are getting popular in the world. One of the most important technical challenges here is to efficiently analyze large graph data that express various types of relationship between people, items, and places. In this talk, we overview the trend of Big Data and IoT and then explain our research on distributed query optimization on cloud environment and efficient graph mining algorithms. Finally, we introduce some of our interesting applications of Big Data: 1) social network analysis by employing graph mining algorithms, 2) business data analysis by exploratory data analysis techniques, and 3) Smart route recommendation system empowered by IoT.

ABSTRACT. Vehicle Routing Problems (VRP) involve designing least-cost delivery routes to visit a geographically-dispersed set of customers. Over the past 60 years, this class of problems has been the subject of considerable work, summing up to thousands of articles. In 2017, we can reasonably say that the classical “capacitated” VRP (with only capacity constraints) is fairly well solved by metaheuristic techniques. Yet, the research on VRPs keeps on expanding even further, as a consequence of the increasing diversity of applications, which bring forth new difficult constraints, objectives, and combined decisions to account for customer’s needs, vehicle and network restrictions, and to better integrate VRP optimization in the decision chains. Moreover, with the advent of collaborative logistics, green initiatives, smart cities, multi-modal transport, in contexts where multiples stakeholders and conflicting objectives have to be considered jointly, or in the presence of dynamic problems with a short response time, the efficient resolution of these problems becomes even more critical. In this talk, we will review some of the most challenging and recent VRP variants, and examine the heuristic solution techniques which are developed to tackle them. We will study the close connections between the structure of the problem decision sets, and the associated solution methods, showing how modern heuristics can effectively perform a search in a reduced space, defined by fewer groups of decision variables. Finally, a key challenge is to progress towards “unified” solution methods, which are not tailored for one single problem, but instead designed to solve a wide collection of problem variants with different constraints and objectives. For this purpose, we expose some of the main principles of the Unified Hybrid Genetic Search (UHGS), which has been recently extended to obtain state-of-the-art results --in a single code base-- for more than 50 difficult variants of vehicle routing and arc routing problems.

ABSTRACT. Paraphrase identification is a very important problem and is used in many natural language processing tasks such as machine translation, bilingual information retrieval, plagiarism detection, etc. With the development of information technology and the internet, the requirement of the textual comparing is not only in the same language but also in many different language pairs. Especially in Vietnamese, the need to detect paraphrase in English-Vietnamese sentences is very large because English is a most popular foreign language in Vietnam. However, the in-depth studies on cross-language paraphrase identification between English and Vietnamese are still limited. In this paper, we propose a method to identify the English-Vietnamese cross-language paraphrase cases using a fuzzy-based method and the BabelNet semantic network. We identify if the pair of sentences is paraphrased by using feature classes and then combine these results into a final one using a mathematical formula. The experimental results show that our model achieves 77.1% F-measure accuracy and has the advantage of the processing speed compared to other methods which have equivalent quality.

ABSTRACT. This paper presents a method for cross-domain intention detection, which identifies posts expressing user intentions in discussion forums. The results of the task can be beneficial for intelligent business such as online advertising and marketing. Our method utilizes labeled data from several other domains to help the learning task in the target domain using a Naive Bayesian framework. Since the distributions of data vary from domain to domain, it is important to adjust the contributions of different data sources in the learning model to achieve high accuracy. Here, we propose to optimize the parameters of the Naive Bayes classifier using a stochastic gradient descent (SGD) algorithm. Experimental results show that our method outperforms several competitive baselines on a benchmark dataset, consisting of forum posts in four domains, namely Cellphone, Electronics, Camera, and TV.

ABSTRACT. Text compression is the task of identifying the main information in the source text to form a short single sentence. A broad approach is to find a path containing common vertices in the word graph model. The first issue of this approach is that the path finding algorithm can separate words from the phrase expressing a content. This leads to create new sentences having different meaning from the original ones. The second issue is that when an information is expressed by different words or phrases, called co-reference situations. Due to lacking of mechanism for handling this situation, the compression will be missing information. We propose in this paper a method to overcome the above issues. The core of new method is the improved graph model in which each vertex illustrates a phrase with its corresponding Part-of-Speech label. The intersection vertices of branches are results of mechanism for handling co-references. The compressing algorithm reduces the graph and forms the final sentence. We use ROUGE measure to compare with two word graph-based baselines. The experiment result shows that our method creates short sentences containing rich information.

ABSTRACT. Discourse* analysis is an important natural language processing task. There are many discourse parsers in many languages, such as English and Chinese, constructing discourse trees from text documents for further semantic analysis. However, there is lack of a Vietnamese discourse treebank for research in Vietnamese discourse parser. Therefore, this paper presents some problems in building discourse treebank and proposes a discourse annotation framework for building Vietnamese discourse treebank. Then, two experiments in discourse relation classification and in discourse nucleus classification with the preliminary result of constructing Vietnamese discourse treebank are presented to show its feasibility.

ABSTRACT. Open information extraction (OIE) is the process to extract relations and their arguments automatically from textual documents without the need to restrict the search to predefined relations. In recent years, several OIE systems for the English language have been created but there is not any system for the Vietnamese language. In this paper, we propose a method of OIE for Vietnamese using a clause-based approach. Accordingly, we exploit Vietnamese dependency parsing using grammar clauses that strives to consider all possible relations in a sentence. The corresponding clause types are identified by their propositions as extractable relations based on their grammatical functions of constituents. As a result, our system is the first OIE system for the Vietnamese language that can generate open relations and their arguments from Vietnamese text with highly scalable extraction while being domain independent. Experimental results show that our OIE system achieves promising results with a precision of 83.71%.

ABSTRACT. Suggestion intents mining from texts is an emerging research topic in recent years. Fully understanding suggestion intents in conversational texts is a complicated process that includes three major stages: user suggestion intents filtering, suggestion domain identification, and arguments extraction of suggestion intents. In the scope of this paper, we study the first phase, that is, building a binary classification model to determine whether a text unit carries suggestion intents or not. We come up with a new text unit to analysis suggestion, that is functional segment. According to the ISO 24617-2 standard, a functional segment is ``minimal stretch of communicative behavior that has one or more communicative functions''. We investigate two approaches to filter functional segment containing suggestion intents: machine learning using maximum entropy (Maxent) model and deep learning using convolutional neural networks (CNN) model. The results of these experiments on Vietnamese online media texts are very promising. To the best of our knowledge, this is the first study to analyze suggestion at functional segment unit.

ABSTRACT. Motion detection in pixel domain of videos often requires highly capable resource to perform. This is due to computational demands for decoding the videos to obtain visual image. The problem can be
solved by performing motion detection directly on bit streams of the compressed video data. This paper presents a new method for moving object detection in compressed domain of high resolution
surveillance videos. In this work, we use video coding parameters (eg. motion vectors) extracted directly from bit stream of the compressed video to separate the moving blocks out of video background. We then propose new techniques for the segmentation and refinement of the foreground blocks to detect the moving objects. Experiments have been conducted on both public and self-recorded video data sets. Experimental results show the outperformance of our proposed method in comparing with conventional one.

ABSTRACT. Abnormal activity detection is an important issue in video surveillance. There are two approaches to classify activity: generative model and discriminative model. Abnormal activity detection is different than detecting activity because of “anomalies” is unpredictable, their appearance is unknown before, so we cannot use discriminative model to detect abnormal based on training data of abnormal activity. One class learning method is used to model normal behaviors by generative model and detect abnormal activities which do not satisfy this model. In this paper, we use fast dense spatial-temporal features within regions of interest (ROIs) to model normal behaviors by Support Vector Data Description (SVDD). Besides, we use KMean++ algorithm to cluster normal data then the multi hyperspheres SVDD are constructed separately on clusters instead of only one hypersphere SVDD on multi-distribution data. Experiments on benchmark datasets contain various situations with human crowds, overlapping between individual subjects and low resolution. The experiments show that our approach is able to detect abnormal behavior more effectively and especially is outperforming state of the art methods on Ped2 dataset.

ABSTRACT. This paper presents a solution to improve performance of human action recognition from multiple camera views. For each camera view, we started by investigating a bag of word model that consists of STIP features to capture motion, a random forest for feature quantization and a SVM for action classification as baseline. However, to avoid background effect, we take STIP features only in the moving regions detected by background subtraction technique. Then, as some actions are very similar (interclass similarity), they discriminate against each others by some minor motions of body part (hand or foot) or/and by the order of movement during the action, we propose a spatial-temporal pooling of STIP features to take this difference into account. Finally we propose a strategy of sifting views in testing phase to deal with difference of camera viewpoints from training phase. The result from each view will be combined by late fusion. The proposed method has been evaluated on a benchmark multi-view dataset WVU. Experimental results show that this method achieves 95.5% of accuracy, that outperforms the baseline by 28.5%. It is evenly better than a method using advanced convolution neural network by 5.5%.

ABSTRACT. We propose a novel method to classify websites based on their functional purpose. A website is classified either as single service, brand or service directory. We utilize a number of features that are derived from the link of the website, the postal addresses found in the website, the size of the website, and the text of the anchor element in the Document Object Model tree. We investigate two models to perform the classification task: decision tree and clustering-based models. Our method is fully automated and does not require extensive training data or user interaction. The proposed website classifier improves the baseline by 2 percentage points in case of single service, 33 percentage points in case of brand and 18 percentage points in case of service directory

ABSTRACT. Facial expression analysis plays a key role in analyzing emotions and human behaviors. Smile detection, emotion recognition and gender classification are special tasks in facial expression analysis with various potential applications. In this paper, we propose an effective architecture of Convolutional Neural Network (CNN) which can jointly learn representations for three tasks: smile detection, emotion recognition and gender classification. In addition, this model can be trained from multiple sources of data with different kinds of task-specific class labels. The extensive experiments show that our model achieves superior accuracy over recent state-of-the-art techniques in all of three tasks on popular benchmarks. We also show that the joint learning helps the tasks with less data considerably benefit from other tasks with richer data.

ABSTRACT. Network traffic classification is an important problem in network traffic analysis. Network traffic classification plays a vital role in many network tasks including quality of service, firewall enforcement and security. One of the challenging problems of classifying network traffic is the imbalanced property of network data. Usually, the amount of traffic in some class is much higher than the amount of traffic in other classes. In this paper, we proposed an application of a deep learning approach to address imbalanced data problem in network traffic classification. We used a recent proposed deep network for unsupervised learning called Auxiliary Classifier Generative Adversarial Network to generating synthesized data samples for balance between the minor and the major classes. We tested our method on a well-known network traffic dataset and the results showed that our proposed method achieved better performance compared to a recent proposed method for handling imbalanced problem in network traffic classification.

ABSTRACT. In this paper, we investigates an elastic optical network that is based on coarse granular routing. The network can exploit both elastic optical networking and coarse granular routing techniques to deal with the trade-off between the link cost and the node cost for creating spectrum-and-cost efficient solution for future Internet backbone networks. Firstly, we have proposed an optical cross-connect (OXC) architecture that is capable of routing bandwidth-flexible lightpaths coarse-granularly. We, then, evaluated spectrum utilization of the elastic optical network employing the developed OXC architecture with typical modulation formats under various network and traffic conditions. Finally, numerical evaluation was used to verify the spectrum utilization efficiency of the proposed network in comparison with that of conventional WDM network and traditional elastic optical network.

ABSTRACT. The field of interconnection network has recently observed a resurgent interest that are fueled by modern ICT technologies wherein traditional topologies have been found less desirable. The recent approach using the random network model has been considered a good answer to new challenges but also has a limitation in routing scalability, i.e. the routing table size (RTS) can become very large which is not affordable by market commodity switches. We propose to design new routing schemes especially designed for the random network model that can improve this situation. As a major contribution, we propose a new compact routing scheme named by $CORRA$ which exploits the random model properties that help to achieve near optimum route length while keeps the routing table size small enough. In general, our new routing algorithm can achieve average route length only a bit longer than in using the shortest path routing. The RTS achieved by CORRA is very small as expected, e.g. 520 for network size 8192, which is small enough to allow network size even up to millions of nodes.

ABSTRACT. In this paper, we will describe a new programming framework call U-RPC to ensure a fault tolerant distributed computing system by using Distributed Hash Table (DHT). DHT is widely used in P2P networks as a scalable mechanism to store and get data since many years ago. Many of the DHT algorithm and implementation was proposed and applied in the real system, each one has advantages or disadvantages depending on the context of the system. In our pro-posal, we would like to add an RPC layer on top of any DHT. With U-RPC, users can do many more complex tasks in the DHT but not limited to traditional tasks: put/get/free anymore. In order to prove the correctness of our framework, we had implemented our framework on top of Kademlia and deployed it on Grid5000, a French large scale, testbed for experiment driven research in all areas of computer science. Our results show that we can easily provide an RPC layer for an existing Kademlia DHT to use in microservices architecture and guarantee the fault tolerance.

ABSTRACT. Multipath TCP (MPTCP) enables concurrent data transmissions over multiple wireless links aiming to achieve bandwidth aggregation and resilience enhancement. MPTCP is considerably software-based, hence it is expectedly evolved and emerged in the current and next generations of mobile wireless networks. The canonical and most important use case of MPTCP in mobile wireless is the scenario, in which a mobile device installed MPTCP uses two wireless links (i.e., LTE and Wi-Fi) for data transferring with an application sever. In this paper, we conduct a study of MPTCP performance in the such scenario, focusing on state-of-the-art implementations (i.e., balia, olia, lia, wvegas). First, we investigate the backward compatibility of MPTCP implementations in a single-path environment. The results show that wvegas is not well fit, while the others are compatible. Secondly, we evaluate the effects of Wi-Fi or LTE selection for the MPTCP initialization on the overall performance. While the handshaking time is similar for all the MPTCPs, the throughput is largely different. The right selection will improve throughput up to more than 200% (i.e., in the case of olia).

ABSTRACT. Dynamic Adaptive Streaming over HTTP (DASH) has been widely used in online video streaming nowadays. To reduce network congestion, popular videos are replicated in the intermediate cache nodes. However, with DASH, the intermediate cache node may lead to bitrate oscillations due to the difference in the throughputs from the end user to the cache node and from that to the origin server. In this paper, we develop an adaptation logic for DASH that can reduce bitrate oscillations. Our proposed adaptation logic combines both throughput-based and buffer-based adaptation methods. Extensive simulations show the efficiency of our proposed method.