Abstract In video streaming, bandwidth constraints significantly affect client-side video quality. Addressing this, deep neural networks offer a promising avenue for implementing video super-resolution (VSR) at the user end, leveraging advancements in modern hardware, including mobile devices. The principal challenge in VSR is the computational intensity involved in processing temporal/spatial video data. Conventional methods, uniformly processing entire scenes, often result in inefficient resource allocation. This is evident in the over-processing of simpler regions and insufficient attention to complex regions, leading to edge artifacts in merged regions. Our innovative approach employs semantic segmentation and spatial frequency-based categorization to divide each video frame into regions of varying complexity: simple, medium, and complex. These are then processed through an efficient incremental model, optimizing computational resources. A key innovation is the sparse temporal/spatial feature transformation layer, which mitigates edge artifacts and ensures seamless integration of regional features, enhancing the naturalness of the super-resolution outcome. Experimental results demonstrate that our method significantly boosts VSR efficiency while maintaining effectiveness. This marks a notable advancement in streaming video technology, optimizing video quality with reduced computational demands. This approach, featuring semantic segmentation, spatial frequency analysis, and an incremental network structure, represents a substantial improvement over traditional VSR methodologies, addressing the core challenges of efficiency and quality in high-resolution video streaming.

Currently, microservices are trending as the most popular software application design architecture. Software organisations are also being targeted by more cyber-attacks every day and newer security measures are in high demand. One available measure is the application of anomaly detection, which is defined as the discovery of irregular or unusual activity that occurs to a greater or lesser degree than normal occurrences in a data series. In this paper, we continue existing work where various real-world cyber-attacks are executed against a running microservices application, and the application traffic is logged and returned in the form of distributed traces. A Diffusion Convolutional Recurrent Neural Network is used to model the set of distributed traces and learn the spatial and temporal dependencies of the application traffic. Subsequently, the model is used to make predictions for ongoing microservice activity and threshold-based anomaly detection is applied to detect irregular microservice activity indicating the presence of seeded cyber security attacks, or anomalies. The cyber-attacks used to evaluate this approach include a brute force attack, a batch registration of bot accounts and a distributed denial of service attack.

Content replication and name-based routing lead to a natural multi-source and multipath transmission paradigm in NDN. Due to the unique connectionless characteristic of NDN, current end-to-end multipath congestion control schemes (e.g. MPTCP) cannot be used directly on NDN. This paper proposes a Network Utility Maximization (NUM) model to formulate multi-source and multipath transmission in NDN with in-network caches. From this model, a family of receiver-driven transmission solutions can be derived, named as path-specified congestion control. The path-specified congestion control enables content consumers to separate the traffic control on each path, which consequently facilitates fair and efficient bandwidth sharing amongst all consumers. As a specific instance, a Delay-based Path-specified Congestion Control Protocol (DPCCP) is presented, which utilizes queuing delays as signals to measure and control congestion levels of different bottlenecks. In addition, a set of high-performance congestion control laws are designed to accelerate bandwidth and fairness convergence towards the optimum defined by the NUM model. Finally, DPCCP is compared with state-of-the-art solutions. The experimental evaluations show that DPCCP outperforms existing solutions in terms of bandwidth utilization, convergence time and packet loss.

Content replication and name-based routing in Named Data Network (NDN) naturally lead to connectionless multi-source and multipath transmissions. Traditional congestion control designed for end-to-end connections cannot well fit this architecture. Explicit congestion notification (ECN) can better support NDN because congestion is detected where it occurs and ECN can timely notify the traffic initiator of congestion. NDN can be deployed as an overlay protocol (sharing the underlying devices with other protocols), which means the congestion may also occur at an underlying device (e.g. a switch). In this case, the NDN nodes cannot access the queue or other link status at a remote underlying device for congestion detection. A promising ECN scheme must be able to detect congestion happening anywhere (at an NDN node or an underlying device) without using underlying link information. This paper proposes Hop-byHop Congestion Measurement (HbHCM) and Practical Active Queue Management (PAQM) to enable detecting congestion and generating ECN at NDN nodes via monitoring the change of transmission delays. HbHCM measures the transmission delay at the hop level and PAQM converts the delay to ECN signals to notify consumers. We compared HbHCM + PAQM with two milestone solutions (router-label and ECN-based). The simulation results show that HbHCM + PAQM can accurately detect congestion, improve bandwidth utilisation and better support multipath transmission, no need to rely on route or link information.

Named Data Networking (NDN) is a promising communication paradigm to support content distribution for the Future Internet. The objective of this paper is to maximize the consumer downloading rate by retrieving content via multiple paths concurrently. This is supported by adaptive forwarding in NDN. The majority of solutions for selecting the forwarding interfaces do so based on latency. However, this can overload the low-latency paths quickly. This can occur as users reduce requesting rates based on congestion signals, from the low-latency paths. Hence, the high-latency paths are not fully utilized. This paper solves this problem by introducing Backpressure Interest Control Protocol (B-ICP). In B-ICP, routers estimate the forwarding capabilities of interfaces based on congestion signals and limit the forwarding rates to interfaces accordingly. Thus, B-ICP avoids congesting certain paths earlier than others, with the aim being evenly distributed path utilization. Simulation-based evaluations show that B-ICP improves throughput, converges to equilibriums quickly and supports the producers that join the network dynamically in comparison with existing solutions.

Named Data Networking (NDN) is a promising Future Internet architecture to support efficient content distribution. Specifically, P2P may gain benefits from NDN, as NDN inherently provides a flexible forwarding plane for multi-source and multi-path communications. Existing studies in this area have proposed solutions, but these are adversely affected by link latency. This leads to illogical resource allocation and low link utilization for P2P. In this paper, we propose a new Heterogeneous-Latency Adaptive Forwarding (HLAF) strategy for peer-assisted video streaming in NDN. In peer-assisted video streaming, users (peers) proactively share the available content to others. By measuring the performance of forwarding interfaces, using both the level of congestion and the round-trip time, HLAF enables efficient P2P communication, which minimizes the latency and enhances the throughput. The experimental results show that the proposed strategy can enhance the peers' Quality of Experience (QoE).

The Internet of Things (IoT) places significant demands on network infrastructure in order to process data captured by ubiquitous sensor devices. One existing technique to support this sensor data processing involves transporting captured data to cloud servers. This approach suffers from numerous issues such as increased transmission costs i.e. bandwidth consumption and delays. To help resolve these issues, this paper proposes Programmable IoT (PIoT), a novel IoT data processing architecture. It is an application layer design which operates over Named Data Networking (NDN) to enable the execution of reconfigurable processing-logic in the network. In addition, a novel naming scheme and computation service for IoT is presented to describe the processing requirements using Lambda Expressions. To verify the feasibility of our design, a real-world implementation was created and evaluated. It compares efficiency of the in-network versus out-network approaches.