3GPP Access Traffic Steering, Switching, and Splitting (ATSSS) enables traffic to be distributed across heterogeneous 3GPP and non-3GPP access networks to improve performance, reliability, and resilience. ATSSS can use multipath transport protocols such as Multipath QUIC (MP-QUIC), where packet scheduling plays a central role in determining latency and resource utilization for delay-sensitive applications. Many existing MP-QUIC scheduling policies rely on instantaneous path measurements or fixed rules rather than forecasts of future application demand. In multi-flow scenarios, such decisions can lead either to contention on the preferred low-latency path and transient latency inflation for priority traffic or to overly conservative use of available capacity. This paper proposes iScavenger, a predictive, machine-learning-based multi-flow scheduling policy for ATSSS environments. iScavenger employs a Long Short-Term Memory (LSTM) model to forecast near-future bandwidth demand for delay-sensitive Sticky traffic. Based on this prediction, background packets are admitted to the preferred low-latency path only when sufficient residual capacity is expected to remain; otherwise, they are steered to the alternative path. The policy is implemented within the Monty MP-QUIC framework and evaluated in a controlled Mininet testbed using traffic traces from the online game League of Legends, with fixed and variable path capacities and controlled jitter and packet loss. The results indicate that, under the evaluated conditions, iScavenger provides configurable operating points in the latency–utilization trade-off, limiting additional Sticky-flow RTT while achieving higher background throughput than conservative baseline policies. These findings highlight the potential of short-term traffic-demand prediction for proactive contention management in ATSSS-enabled multi-access networks.

360° video streaming is rapidly advancing, with emerging applications in immersive communications envisioned for 6G Networks. However, dynamic user VR navigation (e.g., with head-mounted displays) under variable network conditions remains challenging, leading to severe stalls and quality degradation. In this work, we experimentally evaluate 360° video streaming over state-of-the-art protocols such as HTTP over QUIC, i.e., HTTP/3, focusing on the impact of stream scheduling. We implement and evaluate scheduling policies in two classes: (1) explicit priority, coordinated with QUIC, which dynamically prioritises 360° video tile requests based on temporal and spatial criteria; and (2) implicit priority, confined to the application layer and hidden from QUIC, relying on request order strategies. In realistic experiments with network emulation and 360° user navigation, we observe that explicit priority policies applying temporal criteria to urgent tiles reduce stall time by up to 81 % while maintaining competitive video quality compared to application-only stream scheduling.

In all networking systems, queuing is important to ensure appropriate resource utilization in the presence of bursty traffic and varying traffic demands. The Starlink access network is additionally also dynamic in terms of the capacity it can provide, and thus queuing plays an even greater role to ensure appropriate communication performance for the end-users while maintaining high resource utilization. However, for Starlink most system design details, along with the setup of the internal queuing, is private information and not publicly available. To address this we have developed a high-precision, burst-pattern controlled, traffic generation approach allowing us to precisely measure the one-way delay for Starlink. By analyzing the delay and loss in conjunction with a queue simulator we find that Starlink does not employ per-flow fair queuing or drop-tail buffers, but it does use drop-front buffer management. While drop-front reduces delay, it may also interfere with the assumptions made by loss-based congestion controls, potentially contributing to throughput degradation.

6G Internet of Things (IoT) systems will support massive deployments of energy-constrained devices operating over long lifetimes, with sustainable and reliable communication being a key target. Proper design of future IoT is expected to rely on an in-depth understanding of existing cellular IoT (CIoT) deployments, currently based on Long-Term Evolution Machine Type Communication (LTE-M) and Narrowband-IoT (NB-IoT) technologies. However, the impact of operator configuration decisions on device energy-efficiency and quality of service (e.g., transmission latency) remains under-explored and is rarely analyzed across deployments. This paper presents a measurement-based analysis of energy consumption and latency in commercial CIoT deployments across the Nordic countries. Using controlled laboratory measurements, we evaluate Radio Resource Control (RRC) connected-state duration, connected-state energy consumption, and application-layer round-trip time (RTT) under different radio conditions. The results indicate that operator-specific configurations have a key impact on performance differences across LTE-M networks. In particular, short RRC inactivity timers reduce energy consumption by shortening connected-state durations. Additionally, activating the Connected-state Discontinuous Reception (cDRX) improves efficiency during longer connections. In contrast, when cDRX is disabled, energy consumption increases despite similar connection times. For NB-IoT, working in poor radio conditions (i.e., adopting higher coverage enhancement levels) significantly increases connected-state duration, energy consumption, and latency. Overall, LTE-M provides lower energy consumption and latency, whereas NB-IoT extends coverage at the cost of higher delay and energy use. These findings show that operator configurations are a dominant factor for IoT deployment effectiveness and suggest that adaptive configuration mechanisms are important for improving the efficiency and performance of future CIoT systems.

With networking moving into the sub-millisecond latency domain, latency in the end host itself can become a significant barrier to achieving consistently low application latency. Both the physical interconnect between the network card and the CPU, the kernel network stack, and the scheduling of applications themselves can be considerable sources of latency. Previous work has studied host latency at various levels, yet there remains a lack of methods and tools to continuously monitor host latency in production. To remedy this, we present netstacklat, a monitoring tool that captures latency at several points in the host network, from the early parts of the Linux kernel network stack all the way until the application reads the data. We evaluate netstacklat in a testbed, demonstrating its ability to capture host latency across 144 variations of HTTP workloads for Nginx and Apache, while also showing how the low monitoring overhead does not inflate tail latency by more than 6

TCP Prague is a proposed congestion control for L4S, a new Internet architecture. L4S promises very low queuing delay on the Internet. It achieves this by a fundamental change to the meaning of Explicit Congestion Notification (ECN) marks; Active Queuing Management (AQM) should mark packets with ECN a lot earlier and more frequently to better communicate the congestion state to congestion control algorithms. These algorithms use this information to fine tune their sending rate. TCP Prague does so by reducing its congestion window marginally on each marked packet, like Data Center TCP. It would, however, starve classic flows if they were to compete on a non-L4S ECN AQM, as this behavior conflicts with the classic interpretation of ECN marks. To avoid this, Prague tries to detect these cases and fall back to classic behavior to ensure fair flow rate sharing at such bottlenecks. A recent study by Sarpkaya et al. has shown this fallback to significantly hurt the Prague flow in L4S environments. We reproduce the issue and analyze it in different scenarios to highlight the severity of the issue. We then use eBPF to analyze the fallback implementation and find a bug that causes Prague always to fall back to classic behavior, even on an L4S AQM. We then also visualize the sharing characteristics of Prague in different scenarios, with the fallback corrected, and show that Prague now behaves as expected.

The deployment of Cellular Internet of Things (CIoT) is expected to reach over six billion devices by 2030. Many of these devices will be located in remote areas where replacing or recharging their batteries would be difficult and expensive. Therefore, it is crucial to configure these devices for efficient energy use to avoid frequent battery replacements or recharging. However, optimizing the energy consumption of CIoT devices, considering their applications and operating environmental conditions, presents a complex challenge. In response to this challenge, we propose the Gradient-Boosted Learning Optimization for Battery Efficiency (GLOBE) framework for dynamic configuration of Narrowband Internet of Things (NB-IoT) devices. GLOBE adjusts the radio layer of NB-IoT devices based on data transmission patterns and network conditions, enabling swift and automated reconfiguration. Our results demonstrate that GLOBE reduces energy consumption by 30% to 75% compared to baseline configurations, offering significant benefits for both network operators and end devices by improving energy efficiency.

The Transport Services architecture [I-D.ietf-taps-arch] defines a system that allows applications to use transport networking protocols flexibly. This document serves as a guide to implementation on how to build such a system.

Mobile Augmented Reality (MAR) applications pose unique challenges due to computation intensity, constrained device resources, and high interactive rendering requirements. The emergence of 5G and edge computing offers opportunities to offload computation to the edge and cloud, indirectly enhancing the computing capability and usage duration of MAR devices. However, existing general task offloading and multipath transmission techniques do not address the challenges in offloading path selection with multiple edges, dynamic resource competition awareness, and spatial computation with strong task dependencies. This paper contributes FPSelector, a flexible path selector for MAR offloading. We present a two-tier MAR-specific offloading scheme with multiple edge nodes. In offloading decisions, we design a reinforcement learning model to generate the selection policy for each packet of an AR data stream. This model incorporates an action masking mechanism, a comprehensive reward function, and state features complemented by a resource prediction module, making FPSelector aware of dynamic heterogeneous environments. Moreover, we propose an online learning strategy to facilitate real-time selection. To validate its efficacy, we compare FPSelector's performance against leading schedulers under various scenarios, demonstrating a notable reduction of 9.9% and 9.6% in overall completion time for 4 K and 8 K video-based MAR applications compared to its closest competitor.

The promises of multipath transport are to aggregate bandwidth, improve resource utilization, and enhance reliability. In this paper, we demonstrate that the way multipath coupled congestion control is defined today leads to a suboptimal resource utilisation when network paths are disjoint, i.e., they do not share a bottleneck link. With growing interest in standardising Multipath QUIC (MPQUIC), we have implemented the practical shared bottleneck detection (SBD) algorithm from RFC8382 in MPQUIC (MPQUIC-SBD). Through extensive experiments, we evaluate MPQUIC-SBD in the context of video streaming with various Adaptive Bitrate (ABR) algorithms, addressing both ABR classes of rule-based and learning-based solutions. We demonstrate that MPQUIC-SBD accurately detects shared bottlenecks over 90% of the time, depending on the ABR algorithm, as the size of the video segments increases. In non-shared bottleneck scenarios, when MPQUIC-SBD detects that its QUIC subflows do not share the same network resources, it decouples their congestion windows accordingly, enabling video throughput gains of up to 37% compared to MPQUIC. These gains translate directly into improved video quality metrics, including higher bitrate, better resolution, and reduced buffering, resulting in an enhanced quality of experience for users.

The safe and efficient operation of automated vehicles requires processing massive amounts of sensor data. However, the computational capabilities of vehicles are often limited. Recent results point to computational offloading as a promising solution for transferring raw sensor data to be processed elsewhere. This alleviates vehicles from performing costly computations while increasing their perception of complex environments. The work in this paper evaluates the resilience of such solutions, specifically focusing on adverse network conditions, which are often overlooked when evaluating computational offloading. To emulate adverse network conditions, we use synthetic network interference that includes, e.g., packet loss, throughput rate limiting, packet corruption, and RF attenuation. We conducted experiments with a real vehicle on a test track, where object detection was offloaded to an edge server. An optical camera, one of the most common perception sensors, was mounted on the vehicle to scan the environment. The experimental results indicate that network conditions can significantly impact the object detection performance. Packet loss and packet corruption proved to be especially impactful on the accuracy of detections. During the scenario of 5% packet corruption, the median value of false detections reached as high as 20%. The results emphasize the need for resilience and robustness to poor network conditions when designing computational offloading strategies.

Low Earth Orbit (LEO) satellite networks, such as Starlink, are transforming global Internet access by delivering high-speed connectivity to underserved and remote regions. Despite extensive research into Starlink's performance, latency characteristics remain under-explored. This study presents a comprehensive analysis of one-way delay components in the Starlink network using high-frequency, high-precision measurement probes. Over a 10-day period, more than 500 million probe packets were collected and analyzed. The results reveal minor diurnal latency variation and provide means to separate out the delay components contributing to the observed one-way delay, and we sketch a delay model and provide empirical distributions. By measuring both uplink and downlink paths, the study uncovers significant differences in scheduling behavior, with uplink delays more affected by Starlink's periodic 15-second reconfiguration cycles. The results also highlight the limitations of using too coarse measurement intervals, which can introduce aliasing effects. Our OWD data set and traffic generation tool are made available to support further research in the area.

The Narrowband Internet of Things (NB-IoT) technology has an important role in the mobile cellular ecosystem, enabling massive machine-type communication (mMTC) services. NB-IoT propagation was preliminarily analyzed via a measurement campaign carried out in 2020 in the city of Oslo, Norway. This investigation resulted in Oslo-2020, the first NB-IoT-specific alpha-beta-gamma (ABG) path-loss (PL) model, which showed higher prediction accuracy compared to models developed for different technologies but often used for NB-IoT. In this article, to further investigate NB-IoT PL in urban scenarios, we analyze new measurement campaigns performed in 2020-2021 and 2023 in the city of Rome, Italy. First, we use the 2020-2021 measurements to derive Rome-2021, a new NB-IoT-specific ABG PL model. We show that Rome-2021 preserves the statistical properties of Oslo-2020 (e.g., the Gaussianity of the PL exponent distribution across base stations), although the moments of the distributions are different due to city-specific environmental characteristics. We also use new data on signal losses due to outdoor-to-indoor propagation to refine the analysis of this scenario. Finally, we propose a methodology to combine Oslo-2020 and Rome-2021 into a more general model. Our methodology uses so-called mixture distributions (MDs), thus leveraging the shared statistical properties between Oslo-2020 and Rome-2021. By using the 2023 measurements, we show that our MD-based approach estimates PL model parameters with higher accuracy compared to Oslo-2020 and Rome-2021 models used separately, thus providing an effective solution for predicting NB-IoT urban PL in lack of site-specific measurements and information.

Since their deployment and commercialization, 5th Generation (5G) mobile systems have been extensively analyzed to quantify the Quality of Service and Experience (QoS/QoE) achievable by heterogeneous services. Real-time interactive services, i.e., applications within the scope of Ultra-Reliable Low Latency Communication (URLLC) and at the intersection of URLLC and enhanced Mobile Broadband (eMBB), are, however, often tested using simplistic methodologies that do not provide accurate assessments. In this paper, we extend our previous work on the empirical characterization of mobile networks by presenting a comprehensive analysis of a methodology standardized by the International Telecommunication Union Telecommunication Standardization Sector (ITU-T). This methodology is designed for systematic and reproducible QoS/QoE evaluations of real-time interactive services. We validate it through dedicated measurements (for which we open-source the corresponding dataset along with this paper) in the Karlstad University testbed, i.e., a private network supporting 5G connectivity modes and features beyond those available in current public networks in Sweden. Our results, spanning across services, mobility scenarios, connectivity modes, and servers, provide key insights into the intricate dependencies between QoS/QoE, environmental conditions, and system configurations, ultimately serving as a foundation for designing high-performing beyond-5G mobile systems.

We examine the performance of 14 different Linux TCP congestion control (CC) variants over Starlink connectivity. We then focus on the two most commonly used CCs, BBR and Cubic, and the best performing other CC, Illinois. Multiple measurement campaigns were conducted to evaluate throughput performance for single and multiple flows, examine how the performance of the CCs relate to the length of the TCP flow, and investigate how the CCs share resources over Starlink. The results show that BBR (both v1 and v3) outperform Cubic in the current Starlink environment. Cubic, which is the Linux default CC, performs considerably worse than BBR, particularly for single flows. Cubic with HyStart also has additional slowstart performance issues, which can be mitigated by disabling HyStart. The Illinois CC is a promising alternative to BBR and although being less performant for single flows, it has better fairness properties. Leveraging earlier similar measurements for a longitudinal study, we further show that changes made to the Starlink infrastructure during 2023/2024 are likely to have resulted in significantly reduced Cubic throughput.

Existing proposals for positioning in narrowband Internet of Things (NB-IoT) networks based on range estimation are characterized by either low accuracy or lack of compliance with 3GPP standards. While range-free approaches taking advantage of machine learning (ML) have been recently proposed as a potential way forward, their evaluation has been carried out only in simulated environments, with the exception of weighted k nearest neighbors (WkNN), recently tested on experimental data. This work investigates five ML strategies for range-free positioning in NB-IoT networks, based on WkNN and its combination with preprocessing and classification algorithms as well as on artificial neural networks (ANNs). The strategies are evaluated on experimental data and are compared based on a set of key performance indicators measuring both positioning performance and processing load. Two different datasets taken at different times and locations were adopted, enabling the validation of strategies optimized on one testbed on the other, as well as the study of the impact of dataset features on performance. Results show that range-free positioning using ML is a viable solution in commercial NB-IoT networks, and that WkNN and ANNs are at the two extremes in terms of a performance/complexity tradeoff; intermediate tradeoffs can be achieved by combining WkNN with preprocessing techniques and classification models.

Leveraging a data set of almost half a billion packets with high-precision packet times and sizes, we extract characteristics of the bursts emitted over Starlink’s Ethernet interface. The structure of these bursts directly reflects the physical layer reception of OFDMA frames on the satellite link. We study these bursts by analyzing their rates, and thus indirectly also the transition between different physical layer rates. The results highlight that there is definitive structure in the transition behavior, and we note specific behaviors such as particular transition steps associated with rate switching, and that rate switching occurs mainly to neighboring rates. We also study the joint burst rate and burst duration transitions, noting that transitions occur mainly within the same rate, and that changes in burst duration are often performed with an intermediate short burst in-between. Furthermore, we examine the configurations of the three factors burst rate, burst duration, and inter-burst silent time, which together determine the effective throughput of a Starlink connection. We perform pattern mining on these three factors, and we use the patterns to construct a dynamic N-gram model predicting the characteristics of the next upcoming burst, and by extension, the short-term future throughput. We further train a Deep Learning time-series model which shows improved prediction performance.

360 degrees video streaming represents a shift in the way media is consumed: Users can immerse themselves, interact and engage with the viewing environment. This paper proposes the design and implementation of an end-to-end 360 degrees video streaming stack on top of emerging Internet standards such as HTTP/3 and MPQUIC. Preliminary experiments with real network traces show that, compared to single-path streaming, multi-path with stream-aware scheduler and non-blocking buffers achieves up to 25% reduction in tile download times, lower rebuffering ratio (down to 0.53), and decreasing tile misses, while providing a higher Structured Similarity Index Measure (SSIM) of around 17.2 dB.