Mobile Edge Computing (MEC) enables latency-sensitive Internet of Things (IoT) applications by offloading computation to nearby edge servers. However, most existing intelligent scheduling approaches either neglect the full three-tier device–edge–cloud architecture or fail to account for heterogeneous IoT services composed of multiple dependent tasks with diverse deadlines and resource demands. These gaps hinder adaptability under dynamic network conditions. We propose a multi-agent reinforcement learning (MARL) framework in which cellular IoT devices and the edge server act as cooperative agents to optimize task offloading and scheduling. Each agent employs a multi-head deep Q-network (MH-DQN)—with one head per service type—to efficiently manage heterogeneous service workflows. We further implement a multi-head Double DQN (MH-DDQN) variant to improve stability and convergence. In addition, we benchmark our approach against four heuristic baselines and an adaptive simulated annealing (SA) scheduler. Simulation results demonstrate that both MH-DQN and MH-DDQN substantially outperform the heuristic baselines, and that the learned policies match or slightly exceed the SA acceptance ratio while achieving noticeably lower processing times, maintaining high deadline compliance (up to approximately 85% acceptance at heavy load), and reducing latency. MH-DDQN achieves the fastest convergence, greater stability, and slightly lower delays, highlighting its advantage for adaptive scheduling in complex MEC environments.

Power efficiency is a growing concern in softwarized 5G gNBs due to increasing energy costs and carbon emissions. While SmartNICs offer low-power acceleration for packet processing, they struggle with complex operations, often requiring offloading to servers, which raises power usage. This study evaluates the performance and power consumption when dividing 5G gNB data plane tasks between SmartNICs and a DPDK-enabled host, comparing flow-based and function-based task allocation methods. We introduce an adaptive CPU power management strategy that adjusts CPU power states based on traffic. Results show that deploying five SmartNICs with function-based partitioning, which retains packet buffering on the host and offloads header decapsulation, insertion, and lookup-table operations to the SmartNICs, delivers a throughput of 109 MPPS, which is 173% more than a SmartNIC-only setup and reduces latency by 70% compared with a host-only setup. Adaptive power management lowers total power consumption by 12.5% in the optimal partitioning while preserving high throughput and low latency.

Cellular Internet of Things (IoT) offers extensive connectivity today and is poised for further growth in the 5G era, especially after the upcoming sunsetting of 2G and 3G networks. It facilitates crucial IoT applications, such as smart metering to reduce energy consumption, smart logistics to enhance distribution efficiency, and smart environmental monitoring to address urban pollution. To support this expansion, leading mobile operators, global vendors, and developers are deploying NB-IoT networks as part of their long-term 5G IoT strategies. A key goal of NB-IoT is to optimize the battery life of IoT devices. While NB-IoT includes several power-saving features, the cell selection and re-selection processes result in significant energy consumption. We conducted a measurement campaign across three locations in two countries, Norway and Sweden, to investigate this issue based on an NB-IoT commercial network. Our findings reveal that cell re-selection frequently occurs even when the IoT device is stationary. Additionally, the reliance on Reference Signal Received Power (RSRP) for cell selection often leads to oscillations between the nearby cells or prolonged attach procedure. To address this challenge, we propose a cell reselection framework that considers RSRP while also considering historical information on transmission reliability. Evaluations of our proposed framework demonstrate energy savings of over 50% compared to legacy RSRP-based cell selection methods.

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 Cellular Internet of Things (CIoT) has seen significant growth in recent years. With the deployment of 5G, it has become essential to reduce the power consumption of these devices for long-term sustainability. The upcoming 6G cellular network introduces the concept of zero-energy CIoT devices, which do not require batteries or manual charging. This paper focuses on these devices, providing insight into their feasibility and practical implementation. The paper examines how CIoT devices use simultaneous wireless information and power transfer, beamforming, and backscatter communication techniques. It also analyzes the potential use of energy harvesting and power management in zero-energy CIoT devices. Furthermore, the paper explores how low-power transceivers can lower energy usage while maintaining dependable communication functions.

5G mobile networks introduce the concept of network slicing, the functionality of creating virtual networks on top of shared physical infrastructure. Such slices can be tailored to various vertical services. A single User Equipment (UE) may be served by multiple network slice instances simultaneously, which opens up the possibility of dynamically steering traffic in response to the specific needs of individual applications – and as a reaction to events inside the network, e.g., network failures.This paper presents the PoLicy-based Architecture for Network Slicing (PLANS). In this policy framework, the network slice management entity in the 5G core and the UE can cooperatively optimize the usage of the available network slices via policy systems installed both inside the network and on the UE. The PLANS architecture has been implemented and evaluated in a 5G testbed. For two different case studies, we show how such a system can be leveraged to provide optimized services and increased robustness against network failures. First, we consider a drone autopilot scenario, and demonstrate how PLANS can reduce network-slice recovery time by more than 90%. Second, we illustrate for a 360°video streaming scenario how PLANS can help prevent video quality degradation when a network slice becomes unavailable.

Tactile Internet defines applications for remotely controlling and manipulating critical devices that require perceived real-time operation with additional demanding requirements like reliability. These use cases with stringent requirements demand adequate transport protocols to take advantage of the underlying possibilities. Traditional transport-layer solutions like TCP and UDP are no longer sufficient, hence novel protocols are being developed to support these applications. In this paper, we present an implementation and evaluation of the Multi-connection Tactile Internet Protocol (MTIP), a transport layer proposal to support these communications. MTIP uses application and network status information to perform an intelligent selection of paths in order to improve reliability and latency. In our evaluations, we study how the different configurations of the MTIP algorithm affect this selection and we see a direct trade-off where, with more restrictive thresholds, MTIP can increase the packets received correctly at the cost of sending extra duplicate packets.

Cooperative intelligent transport systems (C-ITS) enable information to be shared wirelessly between vehicles and infrastructure in order to improve transport safety and efficiency. Delivering C-ITS services using existing cellular networks offers both financial and technological advantages, not least since these networks already offer many of the features needed by C-ITS, and since many vehicles on our roads are already connected to cellular networks. Still, C-ITS pose stringent requirements in terms of availability and latency on the underlying communication system; requirements that will be hard to meet for currently deployed 3G, LTE, and early-generation 5G systems. Through a series of experiments in the MONROE testbed (a cross-national, mobile broadband testbed), the present study demonstrates how cellular multi-access selection algorithms can provide close to 100 percent availability, and significantly reduce C-ITS transaction times. The study also proposes and evaluates a number of low-complexity, low-overhead single-access selection algorithms, and shows that it is possible to design such solutions so that they offer transaction times and availability levels that rival those of multi-access solutions.

In the fifth generation (5G) mobile networks, the number of user-plane gateways has increased, and, in contrast to previous generations they can be deployed in a decentralized way and auto-scaled independently from their control-plane functions. Moreover, the performance of the user-plane gateways can be boosted with the adoption of advanced acceleration techniques such as Vector Packet Processing (VPP). However, the increased number of user-plane gateways has also made load balancing a necessity, something we find has so far received little attention. Moreover, the introduction of VPP poses a challenge to the design of the auto-scaling of user-plane gateways. In this paper, we address these two challenges by proposing a novel performance indicator for making better auto-scaling decisions, and by proposing three new dynamic load-balancing algorithms for the user plane of a VPP-based, softwarized 5G network. The novel performance indicator is estimated based on the VPP vector rate and is used as a threshold for the auto-scaling process. The dynamic load-balancing algorithms take into account the number of bearers allocated for each user-plane gateway and their VPP vector rate. We validate and evaluate our proposed solution in a 5G testbed. Our experiment results show that the scaling helps to reduce the packet latency for the user-plane traffic, and that our proposed load-balancing algorithms can give a better distribution of traffic load as compared to traditional static algorithms.

By increasing the number of IoT-devices, cloud-computing faces challenges for some computation and time-sensitive applications. Edge-computing has emerged to enable IoT-devices offload their computation tasks. Offloading tasks is a complex and challenging issue. We propose a comprehensive model including user, edge and cloud layers for scheduling continuous offering of services. Furthermore, we modeled the tasks of service as recurrent (repetitive) with a given frequency. The service-placement problem is formulated as a Mixed-Integer Linear Programming problem that aims to minimize the total delay of all services. We solve the problem with CPLEX, and proposed four fast heuristics to find near-optimal solutions. We compared the results of our proposed heuristics with the result obtained with CPLEX, in terms of problem-solving speed and accuracy, as well as resource utilization of all nodes. The results show that two of our proposed heuristics produce near-optimal solutions in a fraction of the time taken by CPLEX.

Cellular networks are continuously evolving to allow improved throughput and low latency performance for applications. However, it has been shown that, due to buffer over-provisioning, TCP's standard loss-based congestion control algorithms (CCAs) can cause long delays in cellular networks. The QUIC transport protocol and the Bottleneck Bandwidth and Round-trip propagation time (BBR) congestion control are both proposed in response to shortcomings observed in TCP and loss-based CCAs. Despite its notable advantages, BBR can experience suboptimal delay performance in cellular networks due to one of its underlying design choices: the maximum bandwidth filter at the sender. In this work, we leverage QUIC's extensibility to enhance BBR. Instead of using the ACK rate observed at the sender side, we apply a more fitting delivery rate calculated at the receiver. Our 5G-trace-based emulation experiments in CloudLab suggest that our modified QUIC could significantly improve latency without any notable effect on the throughput: In particular, in some of our experiments, we observe up to 39% reduction of the round-trip time (RTT) with a worstcase throughput reduction of 2.7%.

Cellular networks have evolved to support high peak bitrates with low loss rates as observed by the higher layers. However, applications and services running over cellular networks are now facing other difficult congestion-related challenges, most notably a highly variable link capacity and bufferbloat. To overcome these issues and improve performance of network traffic in 4G/5G cellular networks, a number of in-network and end-to-end solutions have been proposed. Fairness between interacting congestion control algorithms (CCAs) has played an important role in the type of CCAs considered for research and deployment. The placement of content closer to the user and the allocation of per-user queues in cellular networks has increased the likelihood of a cellular access bottleneck and reduced the extent of flow interaction between multiple users. This has resulted in renewed interest in end-to-end CCAs for cellular networks by opening up room for research and exploration. In this work, we present end-to-end CCAs that target a high throughput and a low latency over highly variable network links, and classify them according to the way they address the congestion control. The work also discusses the deployability of the algorithms. In addition, we provide insights into possible future research directions, such as coping with a higher degree of variability, interaction of CCAs in a shared bottleneck, and avenues for synergized research, such as CCAs assisted by software defined networking and network function virtualization. We hope that this work will serve as a starting point for systematically navigating through the expanding number of cellular CCAs.

The demand for mobile communication is continuously increasing, and mobile devices are now the communication device of choice for many people. To guarantee connectivity and performance, mobile devices are typically equipped with multiple interfaces. To this end, exploiting multiple available interfaces is also a crucial aspect of the upcoming 5G standard for reducing costs, easing network management, and providing a good user experience. Multi-path protocols, such as multi-path TCP (MPTCP), can be used to provide performance optimization through load-balancing and resilience to coverage drops and link failures, however, they do not automatically guarantee better performance. For instance, low-latency communication has been proven hard to achieve when a device has network interfaces with asymmetric capacity and delay (e.g., LTE and WLAN). For multi-path communication, the data scheduler is vital to provide low latency, since it decides over which network interface to send individual data segments. In this paper, we focus on the MPTCP scheduler with the goal of providing a good user experience for latency-sensitive applications when interface quality is asymmetric. After an initial assessment of existing scheduling algorithms, we present two novel scheduling techniques: the block estimation (BLEST) scheduler and the shortest transmission time first (STTF) scheduler. BLEST and STTF are compared with existing schedulers in both emulated and real-world environments and are shown to reduce web object transmission times with up to 51% and provide 45% faster communication for interactive applications, compared with MPTCP's default scheduler.

In this paper, we aim at tackling the scalability problem of the Mobility Management Entity (MME), which plays a crucial role of handling control plane traffic in the current 4G Evolved Packet Core as well as the next generation mobile core, 5G. One of the solutions to this problem is to virtualize the MME by applying Network Function Virtualization principles and then deploy it as a cluster of multiple virtual MME instances (vMMEs) with a front-end load balancer. Although several designs have been proposed, most of them assume the use of simple algorithms such as random and round-robin to balance the incoming traffic without any performance assessment. To this end, we implemented a weighted round robin algorithm which takes into account the heterogeneity of resources such as the capacity of vMMEs. We compare this algorithm with a random and a round-robin algorithm under two different system settings. Experimental results suggest that carefully selected load balancing algorithms can significantly reduce the control plane latency as compared to simple random or round-robin schemes.

Information-centric networking (ICN) with its design around named-based forwarding and in-network caching holds great promises to become a key architecture for the future Internet. Many proposed ICN hop-by-hop congestion control schemes assume a fixed and known link capacity, which rarely - if ever - holds true for wireless links. Firstly, we demonstrate that although these congestion control schemes are able to fairly well utilise the available wireless link capacity, they greatly fail to keep the delay low. In fact, they essentially offer the same delay as in the case with no hop-by-hop, only end-to-end, congestion control. Secondly, we show that by complementing these schemes with an easy-to-implement, packet-train capacity estimator, we reduce the delay to a level significantly lower than what is obtained with only end-to-end congestion control, while still being able to keep the link utilisation at a high level.

Virtualization is central to cloud computing systems. It abstracts computing resources to be shared among multiple virtual machines (VMs) that can be easily managed to run multiple applications and services. To benefit from the advantages of cloud computing, and to cope with increasing traffic demands, telecom operators have adopted cloud computing. Telecom services and applications are, however, characterized by real- time responsiveness, strict end-to-end latency, and high reliability. Due to the inherent overhead of virtualization, the network performance of applications and services can be degraded. To improve the performance of emerging applications and services that demand stringent end-to-end latency, and to understand the network performance bottleneck of virtualization, a comprehensive performance measurement and analysis is required. To this end, we conducted controlled and detailed experiments to understand the impact of virtualization on end-to-end latency and the performance of transport protocols in a virtualized environment. We also provide a packet delay breakdown in the virtualization layer which helps in the optimization of hypervisor components. Our experimental results indicate that the end-to-end latency and packet delay in the virtualization layer are increased with co-located VMs.

This paper aims at increasing our understanding of the behaviour of some of the major TCP variants in LTE networks. A simulation study on the behaviour of seven TCP variants is performed. The evolution of the throughput, congestion window and queuing delay are studied for four scenarios with different network loads and flow types. Our measurements show that, in average radio conditions, most variants are able to quickly reach full link utilisation. However, to achieve the same throughput, they create different amounts of queuing delay. On the one hand, loss-based algorithms tend to completely fill the queue, creating huge queuing delays and inducing packet losses. On the other hand, delay-based variants manage to limit the queue size and decrease the amount of packets dropped by the eNodeB, but struggle to reach the maximum throughput in some circumstances.

There is a growing concern that the Internet transport layer has become less adaptive to the requirements of new applications, and that further evolution has become very difficult. This is because a fundamental assumption no longer holds: it can no longer be assumed that the transport layer is only in the scope of end-hosts. The success of TCP and UDP and the ubiquity of middleboxes have led to ossification of both the network infrastructure and the API presented to applications. This has led to the development of workarounds and point solutions that fail to cover many facets of the problem. To address this issue, this paper identifies requirements for a new transport layer and then proposes a conceptual architecture that we argue is both flexible and evolvable. This new architecture requires that applications interface to the transport at a higher abstraction level, where an application can express communication preferences via a new richer API. Protocol machinery can use this information to decide which of the available transport protocols is used. By placing the protocol machinery in the transport layer, the new architecture can allow for new protocols to be deployed and enable evolution of the transport layer.

To mitigate delay during transmission of bursty signaling traffic, concurrent multipath transmission (CMT) over several paths in parallel could be an option. Still, unordered delivery is a well known problem when concurrently transmitting data over asymmetric network paths, leading to extra delay due to Head-of-Line Blocking (HoLB). The Stream Control Transmission Protocol (SCTP), designed as a carrier for signaling traffic over IP, is currently being extended with support for CMT (CMT-SCTP). To reduce the impact of HoLB, SCTP has support for transmission of separate data flows, called SCTP streams. In this paper, we address sender scheduling to optimize latency for signaling traffic using CMT-SCTP. We present dynamic stream-aware (DS) scheduling, which utilizes the SCTP stream concept, and continuously considers the current network status as well as the data load to make scheduling decisions. We implement a DS scheduler and compare it against some existing schedulers. Our investigation suggests that DS scheduling could significantly reduce latency compared to dynamic path scheduling that does not consider streams. Moreover, we show that naive round-robin scheduling may provide low latency over symmetric network paths, but may transmit data on non-beneficial asymmetric network paths leading to increased latency. Finally, our results show that a static stream based approach, found beneficial for bulk traffic, is not appropriate for bursty signaling traffic.

Currently, Multipath TCP (MPTCP) - a modification to standard TCP that enables the concurrent use of several network paths in a single TCP connection - is being standardized by IETF. This paper provides a comprehensive evaluation of the use of MPTCP to reduce latency and thus improve the quality of experience or QoE for cloud-based applications. In particular, the paper considers the possible reductions in latency that could be obtained by using MPTCP and multiple network paths between a cloud service and a mobile end user. To obtain an appreciation of the expected latency performance for different types of cloud traffic, three applications are studied, Netflix, Google Maps, and Google Docs, representing typical applications generating high-, mid-, and low-intensity traffic. The results suggest that MPTCP could provide significant latency reductions for cloud applications, especially for applications such as Netflix and Google Maps. Moreover, the results suggest that MPTCP offers a reduced latency despite a few percent packet loss, and in spite of limited differences in the round-trip times of the network paths in an MPTCP connection. Still, larger differences in the round-trip times seem to significantly increase the application latency, especially for Netflix, Google Maps, and similar applications. Thus, to become an even better alternative to these applications, this paper suggests that the MPTCP packet scheduling policy should be changed: Apart from the round-trip times of the network paths in a connection, it should also consider the difference in round-trip time between the network paths.

Today commodity mobile devices are frequently equipped with two wireless access technologies, WiFi and 3G/4G. To enable continuous connectivity it is vital that these terminals provide for vertical handover between different technologies. Particularly, they should provide a vertical handover that complies with the timeliness requirements of soft real-time applications. Considering aspects such as cost- and ease-of-deployment, application neutrality, and, not least, the emergence of transport protocols that support multi-homing such as mobile SCTP and multi-path TCP, we think it would be beneficial to handle vertical handover in the transport layer of the mobile terminal. This paper demonstrates through several real-world experiments, the feasibility of using a lightweight vertical handover scheme in smart mobile terminals for live video streaming. The vertical handover criteria is based on the received signal strength. Our experiments suggest that the scheme indeed provides for seamless vertical handover at walking speed - our target scenario. However, the experiments also suggest that the scheme gives significant reductions in handover time, as compared to mobile SCTP without improvements, at higher speeds.

The rapidly growing interest in untethered Internet connections, especially in terms of WLAN and 3G/4G mobile connections, calls for intelligent session management: a mobile device should be able to provide a reasonable end-user experience despite location changes, disconnection periods and, not least, handovers. As part of an effort to develop a SCTP-based session management framework that meets these criteria, we are studying ways of improving the SCTP handover delay for real-time traffic; especially the startup delay on the connection between a mobile device and the target access point. To obtain an appreciation of the theoretically feasible gains of optimizing the startup delay on the handover-target path, we have developed a model that predicts the transfer times of SCTP messages during slow start. This paper experimentally validates our model and demonstrates that it could be used to predict the message transfer times in a variable bitrate flow by approximating the variable flow with a constant dito. It also employs our model to obtain an appreciation of the startup delay penalties incured by slow start during handovers in typical mobile, real-time traffic scenarios.

The stream control transmission protocol (SCTP) is an important component in the ongoing evolution towards IP in the fixed and mobile telephone networks. It is the transport protocol being used in the ongoing deployment of IETFpsilas signaling transport (SIGTRAN) architecture for tunneling of traditional telephony signaling traffic over IP. Further SCTP represents an alternative for future SIP signaling traffic. Key to the success of SCTP is its ability to recover from network failures, in particular failed network paths. SCTP includes multihoming and a failover mechanism which should swiftly shift from a failed or unavailable network path to a backup path. However, several studies have shown that SCTPpsilas failover performance is dependent on factors both related to protocol parameters and network conditions. This paper complements these studies by providing a comprehensive evaluation of the impact of SACK delay under various traffic distributions. The results show a clear relation between the traffic distribution and the impact of the SACK delay on SCTP failover performance. Severe negative effects are observed for low intensity traffic composed of individual signaling messages. On the other hand, our results show limited impact of SACK delay for high intensity and bursty traffic. Furthermore, the results show a limited increase in network traffic by reducing the SACK delay at low traffic intensities and even less impact on network traffic at high traffic intensities. Based on these results we recommend a decrease of the SCTP SACK timer to a small value in signaling scenarios.

The motivation behind this paper is a need to have a more accurate estimation of the failover time in SCTP. The traditional one, commonly used in the literature, is based on the sum of the consecutive retransmission timeouts. This is not always appropriate, especially when using the SCTP multihoming feature as a basis for achieving transport layer mobility in wireless networking scenarios, where the transition time between available paths becomes a key aspect for the optimisation. Two new factors are introduced into the proposed estimation formula to reflect the influence of the network parameters and the behaviour of the most common protocol implementations. For the proposed model, we perform a best-worst case analysis, and then illustrate it with an example of a detailed estimation. Finally, we perform simulations comparing our proposal with the traditional estimation in a typical transport layer mobility scenario including long thin networks.