Future networks with orders of magnitude increase in data rates and large granularity bursty traffic need an architecture with high efficiency and also adapt dynamically to fluctuating offered loads and rapidly changing networks states. Moreover, applications and computing will impose new requirements on the network infrastructure such as time deadlines. The current network management and control systems only adapt quasi-statically (from minutes to days) due the smoothing effects of significant statistical multiplexing of traffic. Future networks will see increase in demands mostly due to large granularity sessions. These granular sessions present large dynamic range and bursty offered traffic to the network, resulting in unpredictable congestions and blocking. We will explore efficient and agile network algorithms to adapt quickly to changing network conditions: a cognitive network management and control system resides in the network control plane as a collection of coordinated algorithms that sense and infer network states, decide and implement fast scheduling of flows, predict intention of users/applications and take appropriate actions, perform rapid load balancing, and handle resiliency via reconfiguration, restoration and reconstitution of failed network assets.

Simultaneous wireless information and power transfer (SWIPT) and cognitive radio (CR) can enable wireless sensor networks (WSNs) overcome the problem of limited service lifetime and the scarcity of spectrum. Cooperative relaying can further improve the performance and capacity of WSNs. This paper mainly investigates the performance of an OFDM based twoway decode-and-forward (DF) relaying cognitive sensor network (CSN) with energy harvesting. The cognitive sensor node serves as an energy harvesting relay which divides the subcarriers into two groups for transmitting the information of the primary users and itself. In order to maximize the transmission rate of CSN, we propose a joint resource optimization algorithm. Simulation results demonstrate the performance of the joint optimization algorithm.

With the development of the Internet of things (IoT), vehicle-to-everything (V2X) plays an essential role in wireless communication networks. Vehicular communications meet tremendous challenges in guaranteeing low-latency transmission for safety-critical information due to dynamic channels caused by high mobility. To handle the challenges, non-orthogonal multiple access (NOMA) has been considered as a promising candidate for future V2X networks. However, it is still an open issue on how to organize multiple transmission links with suitable resource allocation. In this paper, we investigate the problem of the resource allocation for the low-latency NOMA-integrated V2X (NOMA-V2X) communication networks. First, a cross-layer optimization problem is formulated to consider user scheduling and power allocation jointly while satisfying the quality-of-service (QoS) requirements, including the delay requirements, rate demands, and power constraints. To cope with the limited time-varying channel information, a machine learning-based resource allocation algorithm is proposed to find solutions. Specifically, reinforcement learning is applied to learn the dynamic channel information for reducing the transmission delay. The numerical results indicate that our proposed algorithm can significantly reduce the system delay compared with other methods while satisfying the QoS requirements, so as to tackle the congestion issues for V2X communications.

The market for digital services is characterized by strong network effects, positive feedback, and winner-takes-all scenarios. This leads to transient temporal evolution of markets, which cannot be captured using neo-classical economic models. This paper presents dynamic economic market models capturing the temporal evolution of services in the digital economy. In particular, the paper introduces the Buyer-Player-Quitter (BPQ) dynamic market model, in which describes the temporal evaluation of market share in subscription based digital services. Analytical results from the BPQ model are compared to empirical market data for subscription based Massive Multiplayer Online Games (MMOs) in the period 1998-2012. The main finding is that even simple dynamic market models accurately describe the temporal evolution of active players for MMOs.

Virtual network functions, which are placed logically in the path of a network connection, implement important processing steps, such as content inspection, load balancing, etc. While these types of operations are well-understood in the context of the current, connection-oriented Internet architecture, it is unclear if and how such functions can be enabled in emerging architectures, such as Named Data Networking (NDN). In this paper, we argue that virtual network functions can provide benefits to NDN. We present a novel way to use transformations to names in NDN to invoke network functions. We show that network functions can be called either explicitly (e.g., by an end-system) or transparently by intermediate nodes (e.g., by an edge router). We believe that enabling virtual network functions significantly expands the capabilities of Named Data Networking.

Following the rollout of the first 5G networks in 2018, press reports in the US began to emerge that the 5G icon on smartphones was not depicting 5G connectivity. Such reports about `fake' 5G icon reverberated across the industry, exposing a mismatch between the icon on the phone and the actual experience of users. Between 2018 - early 2020, 3GPP and the GSMA sought to provide industry guidance on what and when the 5G icon should be used and how 5G performance can differ from 4G. In this paper, we introduce an intuitive four stage investigation framework to explore the technical considerations that ultimately confirm the veracity of the 5G connectivity. Then, following the launch of 5G in the UK in late 2019, we set out to explore if there were similar confusion on 5G notification and performance in the country. We conducted field measurements at the five busiest train stations in the UK, during rush hour, - using a Samsung 5G S10 and a Samsung S6 Edge+ 4G device - to compare 5G notifications and perceived network performance on 4G and 5G networks. We observe confusing messages to the user - device icon says 5G but Android's TelephonyManager API says 4G; worst cases for latency and uplink/downlink speeds were
minimised but best case performance was the same on 4G and 5G devices. Based on our observations, and while we expect any lingering concerns to be ironed out as 5G deployment and adoption matures, we draw lessons that should guide the industry to avoid doubts about the icon and connectivity in 6G.

LoRaWAN is a long-range, low-power wireless transmission medium, typically used in Internet of Things(IoT) deployments for long range transmission of sensor data and commands for actuator devices. Named Data Networking (NDN) is an emerging information-centric networking architecture with a design suited for handling communication in wireless environments. This work leverages both technologies to create an ad hoc wireless network by creating an application that is able to send and receive NDN packets between LoRaWAN devices without using any IP address. Experimental results gathered in a real environment show that this newly created ad hoc network is capable of transmitting data over distances of at least one kilometer per hop, using NDN features for handling routing and security of data packets. This ad hoc network is suitable for fetching IoT sensor data, sending commands to IoT devices, and for sending small messages between distant nodes in a low-resource, wireless IoT environment with minimal network infrastructure. Furthermore, this system has been used and evaluated while developing an NDN-based IoT sensing platform with built-in security features, such as data signature verification and certificate deployment.

The satellite-terrestrial integrated network (STIN) can realize genuine ubiquitous communications via integrating the satellite network and the terrestrial network. However, the time-varying links and failed nodes in STIN will cause the network connectivity problem. To ensure reliable network connections and efficient integrated communications, the gateway placement issue in STIN should be studied. In this paper, an intelligent gateway placement scheme is proposed to achieve the higher reliability and lower latency in STIN. Firstly, a network model with satellites and terrestrial nodes is constructed. Then, the network reliability is defined by considering the probability of traffic distribution and network connectivity, and the gateway placement problem is formulated to maximize the network reliability under the constraint of propagation latency, which is obtained by path selection via Dijkstra algorithm. Following that, an intelligent gateway placement (IGP) algorithm with the discrete particle swarm optimization (DPSO) named DPSO-IGP is proposed to solve the gateway placement problem. In DPSO-IGP, the improved K-means method is given to reduce the search scope, and the gateway locations with maximum reliability are determined in virtue of DPSO. Finally, simulation results demonstrate that the proposed DPSO-IGP has superiority in terms of the network reliability and latency compared with the existing gateway placement schemes.

This talk does not have an abstract.

Blockchain, the technology behind Bitcoin, has emerged as a “secure by design” technology, showing great potential to enable a wide range of decentralized secure applications across a broad spectrum of industries. However, the security of blockchain is largely depending on a threshold assumption. Using the Bitcoin network as an example, a fundamental assumption is that the network is secure as long as adversaries cannot control more than 50% of the network’s gross computing power. In this talk, we will focus on the proof-of-work (PoW) blockchain consensus protocols and examine some fundamental security properties of blockchain. We show that while blockchain promises decentralization, irreversible record keeping, public verifiability, transparency, and user anonymity, etc., these security properties do not come for free and they are not guaranteed. Excessive overhead and performance deficits may place a fundamental limit on the use of blockchain in many applications. We also show that factors, such as mining strategies and network connectivity, could significantly reduce the fundamental 50% threshold value of blockchain security.

As more and more mobile data is collected continuously in space and time for a variety of purposes such as health monitoring and driving behavior tracking, people are increasingly concerned about their privacy when sharing their data. To minimize privacy leakage, data owners often want to restrict the access to their data based on space and time constraints while meeting each data user's specific needs. In this paper, we introduce an access control system over Named Data Networking (NDN) that allows data owners to schematize and control data access at a fine granularity with respect to time, space, and user needs. More specifically, we designed spatio-temporal access control policies using hierarchically structured names, extended Name-based Access Control to support spatio-temporal policies, and incorporated publish-subscribe functionality for real-time data sharing. Moreover, we developed an NDN-based prototype based on our design and evaluated its performance in sharing both real-time and historical data.

With the rapid growth of mobile video traffic, a wide range of third-party entities need to better understand the status of video quality for QoE monitoring and network management. Video resolution, as a direct reflection of video quality, is a key influence factor of QoE. However, the existing approaches cannot accurately identify the resolution of video traffic under HTTP/2 due to the multiplexing mechanism of HTTP/2. To address this issue, we proposed a method, dubbed as H2CI, to consider the size of the mixed data calculated based on HTTP/2 features as the fingerprint for identification. H2CI can accurately restore the original length of chunks from encrypted video streaming and perform the fine-grained resolution identification. The experimental results show the promise of our method, yielding more than 99% accuracy for fine-grained video resolution identification. Our method can be effectively applied to infer the adaption behavior of video streaming and monitor the QoE of mobile video services under HTTP/2.

With the development of network technology, more and more protocols and devices are used in DDoS reflection and exploitation attacks. Different DDoS attacks often require different responses, so in order to protect against DDoS attacks, it requires not only DDoS detection, but also the classification of the detected DDoS traffic. As DDoS attacks have a wide range of potential targets, the existing algorithms for DDoS detection needs global processing of traffic data from different sources, but such data collection is obviously not conducive to the privacy protection of source users. Therefore, we designed our algorithm for DDoS detection and classification based on federated learning. Deferent terminals only need to pass model gradient parameters rather than directly interact the collected data of them, which can not only reduce communication costs, but protect privacy as well. Considering that network structure should not be too complex for high-rated traffic classification, we use a simplified residual network with fewer parameters for detection and classification.

Multipath BGP (M-BGP) allows a BGP router to install multiple 'equally-good' paths, via parallel inter-domain border links, to a destination prefix. M-BGP differs from the multipath routing techniques in many ways, e.g. M-BGP is only implemented at border routers of Autonomous Systems (ASes); and while it shares traffic to different IP addresses in a destination prefix via different border links, any traffic to a given destination IP always follows the same border link. Recently we studied Looking Glass data and reported the wide deployment of M-BGP in the Internet; in particular, Hurricane Electric (AS6939) has implemented over 1,000 cases of M-BGP to hundreds of its peering ASes.

In this paper, we analyzed the performance of M-BGP. We used RIPE Atlas to send traceroute probes to a series of destination prefixes through Hurricane Electric's border routers implemented with M-BGP. We examined the distribution of Round Trip Time to each probed IP address in a destination prefix and their variation during the measurement. We observed that the deployment of M-BGP can guarantee stable routing between ASes and enhance a network's resilience to traffic changes. Our work provides insights into the unique characteristics of M-BGP as an effective technique for load balancing.

Against the background of the contiguous growth of the Internet and data center traffic, the performance requirement for software middleboxes is increasing rapidly. Although their performance has been improved by continuous research and development, their packet forwarding architecture depends on the CPU and the main memory. Thus, their throughput cannot exceed the limit of the memory bandwidth, for instance. To address these limits, we propose a novel packet forwarding architecture called P2PNIC. In P2PNIC, a NIC directly communicates with other NICs through the PCIe interconnect without CPU and main memory involvement, like the inter-linecard communication in a hardware router. To show the feasibility and the performance advantages of P2PNIC architecture, we implemented P2PNIC on a programmable 40 GbE NIC and compared the throughput and latency with TestPMD, which is an application of DPDK. The evaluation shows that P2PNIC achieves 40.37 Mpps for 64-byte packets, which is 1.45 times higher than TestPMD. In addition, P2PNIC shows 36% lower latency than TestPMD for 64-byte packets with 1 Gbps background traffic. The P2PNIC architecture accelerates packet forwarding on a general-purpose server and advances software-based network technologies.

Managing Quality of Experience (QoE) for users streaming videos over cellular networks (e.g., 4G LTE networks) is an important problem. In this paper, we consider the problem of streaming multiple variable bit rate (VBR) video streams to users over heterogeneous cellular networks (HetNets) comprising of a cellular infrastructure and storage-enabled femtocells that are capable of caching content. We assume that each user can be connected to multiple femtocells simultaneously with the wireless channel quality between users and femtocells varying spatially as well as temporally. We address the joint assignment and scheduling problem, where the assignment problem is to match users to femtocells from where the video can be downloaded and the scheduling problem is to fairly allocate time slots to the different users that are assigned to the same femtocell. Our objective is to maximize the total bit rate received by the users considering all femtocells, subject to the QoE-based video streaming constraint that the number of stalls received by the users is bounded. We formulate the joint assignment and scheduling problem as an optimization problem and develop QoE-aware greedy algorithms that tackle the assignment and scheduling problems separately. We conduct trace-based experiments using real-world 4G LTE wireless connectivity and VBR video traces and demonstrate that our solution outperforms baselines in terms of average number of stalls per user (48% less on average) and results in a fair distribution of stalls across all users.

Application-aware routing has been proposed to facilitate fine-grained control and better resource utilization in Wide Area Networks (WANs). However, there are two key challenges that impact the deployment of application-aware routing in a traditional network: (1) The run time identification of applications and (2) Routing mechanism that is specific to an application's need. However, recent advancements in SDN technology and programmable switches paved a way towards feasible implementations of application-aware routing, by providing flexibility in data-plane for parsing; and in control-plane to employ appropriate routing algorithms.

In this paper, we propose to use the IPv6 flow label for the identification of the applications at run time. We also present a link-weight-based routing mechanism that adapts itself for the traffic of different application categories and results in better throughput (>2x) as compared to the shortest path routing.

We argue that the use of the IPv6 flow label for application identification also does not require any modification at the protocol and switch level in OpenFlow/P4 supporting hardware.

In the video transmission of critical urban events supported by VANETs, intermittent connections, delays, packet losses, and the number of vehicles in collaboration naturally impair the dissemination of data, and hence affect the quality of received video at the destination. Crowdsensing solutions become a real option in helping to handle these events. In this paper, we evaluate the impact of the characteristics of urban VANETs on the human perception of the spatio-temporal context from monitored videos of critical events. We use the MINUET system to assess the quality of videos disseminated and delivered to city authorities, using QoS and QoE metrics, and under different traffic scenarios. Simulated results indicate the system achieves quality levels from satisfactory to good in most of the evaluated scenarios, i.e. average PSNR values between [25-40]dB and average SSIM values between [0.5-1]. In addition, even in adverse situations for detecting the event, video was delivered with a level of quality sufficient to recognize the event.

As a technique to reduce the web response time, HTTP/2 and HTTP/3 have been standardized. However, parallel download in these techniques is possible only for web objects which are provided from the identical object server, so many web objects need to be downloaded from a small number of object servers to improve the effect of HTTP/2 and HTTP/3. To deliver web objects, CDNs delivering objects from cache servers located near requesting users have been widely used. Hence, if we define the co-occurrence degree as the number of webpages in which an object set is included, it is desirable to replace cached objects so that object sets with high co-occurrence degree are remained in the cache. However, the effectiveness of replacing objects based on the co-occurrence degree strongly depends on how the co-occurrence phenomenon of objects appears in the actual webpages. Therefore, in this paper, to clarify the potential of cache replacement based on the co-occurrence degree, we investigate how much the co-occurrence phenomena of objects are observed in actual webpages by measuring the most popular 7,604 webpages. We confirm that the distribution of co-occurrence degree of objects obeys a power law, and the co-occurrence degrees of about 0.01% two-objects sets and 0.001% three-object sets exceed 500. Hence, we confirm the potential of cache replacement based on the co-occurrence degree of objects.

While there have been network studies of traditional network games and streaming video, there is less work measuring cloud-based game streaming traffic and none on Google's Stadia. This paper presents experiments that provide a first look, measure Stadia game traffic for several games, analyzing the bitrates, packet sizes and inter-packet times, and comparing the results to other applications. Results indicate Stadia, unlike traditional network game systems, rapidly sends large packets downstream and small packets upstream, similar to but still significantly different than video and at much higher rates than previous cloud-based game systems or video.

Public Wi-Fi hotspots are essential in most venues and carry over 50% of last-mile Internet traffic for mobile users. Venue owners and mobile advertisers are employing various technologies to promote Wi-Fi monetization via broadcasting advertisements to users. In this paper, we study Wi-Fi advertisement scheduling by considering the communication constraints of Wi-Fi hotspots and the users' tolerance for viewing advertisements. We build a tripartite Wi-Fi advertising model with venues, advertisers and mobile users, and then formulate the Wi-Fi advertisement scheduling process as an optimization problem to maximize users' overall interest value. The formulated problem is a complex mixed integer programming problem. We prove that its objective function is a monotone submodular function subject to a knapsack constraint and a partition matroid constraint, respectively. On this basis, we propose a novel greedy swap algorithm to solve the advertisement scheduling problem with an approximation guarantee of (1-e^(-2))/2 to the optimal solution. Extensive simulation results validate the superiority of the proposed algorithm which obtains the highest user interests towards all displayed Wi-Fi advertisements, thereby enhancing the potential revenues for both venues and advertisers.