The rapid development of Large Language Model (LLM) agents has facilitated the advancement of multi-agent systems, where cognitive knowledge sharing is crucial for the execution of complex tasks. However, achieving the synchronization of the cognitive Knowledge base (KB) among agents under restricted wireless resources remains a challenge, especially in dynamic real-time environments. Therefore, we propose a hierarchical LLM agent system that consists of a high-level Cluster Brain (CB) and multiple Lower-level LLM Agents (LLAs). The cognitive KB of each LLA is represented in the form of a Knowledge Graph (KG). To improve the efficiency of transmitting cognitive KB updates from LLAs to CB, a KG compression framework named MED-EmPress is proposed, which adaptively compresses the semantic features of cognitive KB by applying dimensionality reduction and binary quantization, and then a joint optimization problem of Semantic Compression and Resource Allocation (SCRA) is formulated to maximize semantic fidelity of the cognitive KB being transmitted. To solve this problem, a hierarchical SCRA algorithm is designed to decouple the SCRA problem into two subproblems, which involve dynamically allocating wireless resources and rationally choosing the semantic compression ratio of cognitive KB. The goal is to maximize the system’s semantic fidelity. The evaluation results demonstrate that the MED-EmPress framework reduces the size of the cognitive updates that need to be transmitted by 96%, with only a loss of 3.6% in the entity alignment task. Furthermore, the proposed adaptive compression and transmission scheme improves semantic fidelity by 94% compared to existing methods when wireless resources are severely limited.

The Satellite-Terrestrial Integrated Network (STIN) is a key paradigm to achieve global coverage and ubiquitous connection in the 6G era. Integrating the network slice technology based on Software Defined Networking (SDN) and Network Function Virtualization (NFV) into STIN is recognized as an effective solution to achieve a rapid flexible service provisioning. Specifically, the Content Delivery Network (CDN) service, which is storage resource intensive, is suitable to be deployed on STIN to provide a global range content service suppply. Most existing research on the resource deployment of STIN focuses on static requests, ignoring the dynamic changes in requests caused by the high-speed movement of satellites. Especially, the reconfiguration of CDN slices and the variation of STIN are asynchronous in different time scales: this makes the reconfiguration of CDN slices to cope with faster changing user requests a challenging task. In this paper, we adopt a periodic reconfiguration strategy to configure CDN slices on the edge LEO satellite network of STIN in discrete time intervals. Within each time interval, we formulate the reconfiguration optimization problem to cope with the dynamically changing user requests, wherein the Stochastic Network Calculus (SNC) is used to measure the deployment performance within this time interval. Then, we describe the optimization problem as a Decentralized Partially Observable Markov Decision Process (Dec-POMDP), and propose a reconfiguration algorithm based on Multi-agent Deep Reinforcement Learning (MADRL) to determine the optimal adjustment strategy. Finally, intensive simulations are implemented to verify the performance of the algorithm. Compared with the baselines, the QoS of the proposed algorithm is increased by about 5.8%, while the operation cost and reconfiguration cost are decreased by about 8.5% and 35.3% respectively.

Mobile service continuity is essential important to ensure that user sessions and services will survive user mobility. The 5G enhances its mobility management by providing the flexibility and offering three types of Session and Service Continuity (SSC) modes to address various service continuity requirements. Multi-access edge computing (MEC) is a type of widely adopted network architecture that delivers network services from the boundary of the mobile network by provisioning a set of edge servers. Determining an optimum planning of MEC edge servers, which involves determining edge servers appropriate geographical positions and their serving areas, is a precondition for more efficient service provisioning and better usage of network resources. In this work, we investigate the MEC servers planning problem by considering the management cost for maintaining MEC service continuity. The problem is formulated as a graph partitioning problem to partition the RAN graph with minimum SSC management costs and balanced MEC servers workloads. Then, we adapt a generalizable approximate Graph Partitioning framework which leverages on Graph Neural Network (GNN) to embed the RAN network spacial feature and on Multilayer Perceptron (MLP) for graph partitioning. Based on the framework, we propose a MEC server planning algorithm named MECP-GAP. Finally, we evaluate MECP-GAP with extensive simulations and real network data. Comparing to several baselines, MECP-GAP achieves better performance with lower running time.

The knowledge on a Service Function Chain’s (SFC’s) resource requirements is an indispensable prerequisite for proactive resource provisioning and run-time management of the SFC. However, due to the intrinsic dynamics in network environment, accurate resource requirements and workloads prediction for the Virtual Network Functions (VNFs) of a SFC, especially in a large time scale, is a non-trivial challenge. In the literature, existing works largely neglect the application-level relationship of VNFs in improving prediction accuracy, and few work investigates the multi-step prediction. In this work, we propose a deep-learning-based multi-step prediction model for accurate workload prediction for SFC VNFs in a dynamic network environment. We first demonstrate that predictability can be improved by taking into account application-level dependency by calculating the spatial conditional entropy of adjacent VNFs workloads. Then, the prediction model, named Graph Transformer Networks and sequence-to-sequence LSTM with Attention (GTN-LA) is introduced, which utilizes the Graph Transformer as the encoder to capture the application-level dependencies among VNFs, and the LSTM with attention as the decoder to extract the temporal dependencies within the time varying load information. Finally, GTN-LA is validated through intensive evaluation with a real SFC workload dataset by comparing towards several baselines.

The ability of predicting state of a Network Slice (NS) is indispensable for the run-time management of NSs for providing proactively adjustment and reconfiguration of the NS to avoid Service Level Agreement (SLA) violation and ameliorate network resource utilization. In the literature, NS state prediction methods neglect the spatio-temporal correlation among NS entities. Moreover, NS state involves both Virtual Network Function (VNF) state and transmission link state in the virtual network of the NS. In this paper, we propose an end-to-end model by integrating Graph Neural Network (GNN) and Long Short-Term Memory (LSTM) for the dynamic NS state prediction. Typically, we apply two types of GNN models, Graph Convolutional Network (GCN) and Message Passing Neural Network (MPNN), for aggregating the spatial features for VNFs and transmission links. Then, LSTM is utilized for sequential NS state prediction. Finally, we conducted intensive simulation to validate the effectiveness of the proposed model by comparing to several baselines.

Service Function Chain (SFC) is composed of an ordered set of service functions, which are also noted as virtual network function (VNF) instances, to adaptively form a composite network service with increased flexibility and agility. One fundamental challenge in SFC run-time management is the accurate prediction of the resource load of SFC VNFs for proactive VNF resource provisioning. In the literature, existing works largely neglect the application level relationship for the fine-grained atom-VNFs in microservice architecture. In this work, we propose an end-to-end deep-learning-based prediction model, namely granularity-captured Graph Attention LSTM Network (GGAL), for accurate load prediction at the fine-grained atom-VNF level. By integrating granularity-captured graph attention layers, LSTM and MLP, the model is able to extract the application level spatio-temporal relationship among atom-VNFs and perform accurate load prediction. The effectiveness of the proposed GGAL model is demonstrated through intensive simulations by comparing to several baselines.

This paper investigates a robust resource allocation for reconfigurable intelligent surface (RIS) aided vehicle-to-everything (V2X) communications with imperfect channel state information (CSI). To satisfy the diverse quality-of-service (QoS) requirements of V2X communications, we aim at maximizing the sum capacity of cellular user equipments (CUEs) while guaranteeing the outage probability constraints of vehicular user equipments (VUEs). Then, the considered problem is decomposed into the subproblems of power, spectrum and RIS phase shift op-timization. A graph-based power allocation method is presented to transform the non-convex power allocation subproblem into a tractable one and obtain the closed-form solutions. A worst-case conditional value-at-risk (CVaR) approximation-based method is developed to convert the RIS phase optimization subproblem into a convex semidefinite programming (SDP) problem. We propose a low-complexity learning-based alternating optimization approach which alternately optimizes three subproblems to obtain a near-optimal solution. Simulation results demonstrate that the proposed approach outperforms other benchmark methods.

In the 3GPP-based cellular vehicle-to-everything (C-V2X) architecture, the Roadside Units (RSU) plays an important role for the enhancement of Quality of Service (QoS) of the vehicular applications. The placement of RSUs has been studied in the literature. However, existing works assume known road traffic distribution with given task demands, which is a simplification of the complex real world situation. In this work, we investigate the optimum RSU placement for C-V2X with uncertain traffic density and task demands. We formulate this RSUs Placement in C-V2X Network (RPCN) problem to minimize the expected vehicle tasks offloading delay through uncertain programming where vehicles positions and tasks are treated as arbitrary stochastic variables. We propose a learning-based algorithm by integrating Stochastic Simulation (SS), Artificial Neural Network (ANN) and meta-heuristic algorithm to determine the placement from real traffic data. The proposed method is an offline design with high practicability. We conducted intensive real-trace driven simulations to demonstrate the effectiveness of our approach on placing RSUs with lower task offloading delay.

The emergence of Industrial Internet of Things along with fog computing (FC) has brought great benefits in the industry field through real-time monitoring, resource optimization configuration, and intelligent cloud control. The deployment of fog nodes in industrial plants is the essential precondition for FC implementation. Existing works are mainly based on pregiven sensor traffic in the deployment of fog nodes, which largely ignores the stochastic uncertainty imposed by the dynamic industrial environment. In this article, without requiringa prioriknowledge of the sensor traffic pattern, we establish a mathematical model based on uncertain programming to formulate the fog node location determination and the sensor association problem. Owing to the complexity of the model, we introduce a learning-based algorithm to effectively solve the problem with an acceleration mechanism. Finally, intensive simulations are implemented to verify the performance of the algorithm. The results indicate that our approach outperforms other benchmarks in terms of transmission energy on planning fog nodes in industrial plants with stochastic sensor traffic.

On one hand, the congestion notification mechanism Explicit Congestion Notification (ECN) can only provide coarse-grained congestion signal, which is not sufficient to indicate accurate network and congestion status. On the other hand, the emerging learning-based intelligent congestion control and route selection mechanisms require fine-grained network states information to take accurate actions. This calls for the development of enhanced ECN mechanism to provide precise congestion information and network states. In this work, we design Accurate-ECN, an enhancement of ECN with Inband Network Telemetry (INT) to collect and report detailed network congestion states by attaching network state metadata to the data packets and send back to the sender through TCP ACK by the packet receiver. We designed the Accurate-ECN frame format and the data packet parsing process, and implement the mechanism through the P4 language. Finally, through evaluation, Accurate-ECN is demonstrated to provide various precise network states under different congestion levels.

Digital immersion via Virtual Reality (VR) and Augmented Reality (AR) applications is expected to be a key driver of growth for the 5G mobile network. The immersive requirement imposes many technical challenges. On one hand, Mobile Edge Computing (MEC) is an effective network paradigm to provide low transmission latency and massive computation for 360° videos. On the other hand, viewport adaptive streaming also provides a bandwidth efficient solution. Accordingly, in this paper, we investigate the caching policy for tile-based 360° videos in an MEC caching system. Our goal is to find the optimal caching policy to co-optimize users’ quality of experience (QoE) and MEC energy consumption with no a-priori knowledge on video content popularity. We apply the combinatorial multi-armed bandit (CMAB) theory to solve the above problem which is a sequential decision making problem. On the basis of the combinatorial UCB (CUCB), an improved algorithm is proposed to speed up learning process. The outcome of the algorithm is the caching decision for each time slot. The effectiveness of the proposed learning based caching policy is confirmed by simulation results in terms of the learning speed, hit rate, energy consumption and request latency.

The softwarization and virtualization based Network Slicing (NS) technology provides the momentum for integrating the Information-Centric Networking (ICN) systems into the 5G infrastructure, such that ICN can be virtualized as a NS to co-exist with other IP-based vertical services slices. The implementation of the ICN network slice (ICN-NS) is essentially a 5G end-to-end (E2E) NSs embedding problem, which is normally solved by embedding the given virtual network (VN) of the slice instance. However, determining the details of the VN is non-trivial and largely ignored in the literature. In this work, we formulate the ICN network slices determination and embedding (ICN-NS-DE) problem through an Integer Linear Program (ILP) formulation, such that the ICN-NS determination and embedding problems are jointly solved for a hierarchical ICN system without requiring a-priori knowledge on the VN's topology and resource provisioning information. Due to the complexity of the model, we design an heuristic algorithm for solving the problem in practical large scale network. Finally, we demonstrate the performance of the ICN-NS-DE model and the algorithm through intensive simulations.

Software defined networking (SDN) and network function virtualization have emerged as a promising solution for elastic, dynamic, and scalable network management. A collection of works have investigated how virtualization and cloud technology can ameliorate the infrastructure management for the fifth generation mobile network. However, employing these techniques in mobile network for live streaming has not received enough attention. By leveraging the development of SDN and virtualization techniques, mobile live streaming service providers can efficiently manage their system to cope with network dynamics incurred by the variation of the environment. Specifically, we dynamically instantiate network entities at appropriate locations in response to the user live streaming demands. These network entities are in charge of transcoding and transmitting the live videos to the mobile end users, which we name as virtual live streaming agent (vLA). In this paper, we investigate the dynamic virtualization and migration of vLAs for mobile live streaming services. Typically, rate adaptive streaming is adopted to improve spectrum efficiency and user quality of experience. We formulate an integer linear program for the optimal vLA placement problem. Furthermore, we design a practical SDN-based live vLA migration process. Then, by designing and integrating migration cost functions, we develop a dynamic vLA placement mechanism for dynamic environments with the consideration of migration cost. Both heuristic and on-line algorithms are designed for the vLA placement and migration problems. Finally, a large scale real trace-based simulation is conducted to demonstrate the performance of our vLA placement and migration algorithms.

In a Mobile-CDN, Base Stations (BSs) are equipped with storages for replicating content, and they are allowed to cooperate in replying user requests through backhaul links. In this paper, we investigate the joint optimization problem of content placement and user request redirection for such a BS-based mobile CDN system. Specifically, each BS maintains a transmission queue for replying user requests issued from other BSs. Due to the limited link capacity and the dynamic network environment, the optimization problem should be jointly considered with the transmission queue states. We employ the Stochastic optimization model to minimize the long-term time-average transmission cost under content availability and network stability constraints. By applying the Lyapunov optimization technique, we transform the long-term problem into a set of linear programming (LP) problems, which are solved in each short time duration. Further, we propose a semi-distributed online algorithm to jointly decide content placement and user request redirection. The evaluation confirms that our solution guarantees network stability comparing to the traditional user request redirection scheme.

We consider a mobile-CDN system where base stations (BSs) are equipped with storage for replicating and distributing content. In such a system, BSs cooperation in replying user requests is a widely adopted mechanism. For such cooperative caching, a key issue is the joint optimization of content placement and request redirection, which has been intensively investigated in the literature. However, optimizing this problem to guarantee delay has not received enough attention. Practically, each BS maintains a transmission queue for replying requests issued from other BSs. We investigated the management of such transmission queues to guarantee the queuing delay. By solving an admission rate determination problem and a typical content placement and request redirection problem, the throughput of the system is optimized with no violation on the queuing delay. Finally, a real trace based evaluation demonstrates the benefits of managing transmission queues in improving the mobile-CDN system performance.

In this paper, dynamic resource management is investigated for serving on-demand video streaming users in heterogeneous wireless networks (HWNs) with time varying channel conditions. The HWN is equipped with multi-homing capability, simultaneously connecting to different wireless interfaces. In order to take advantage of the time varying nature of wireless channels, we utilize the joint quality selection associated with quality adjustment at application layer and resource allocation associated with power allocation, subcarrier assignment, and time fraction determination at physical layer to perform the dynamic resource management. By using Lyapunov optimization technique, we develop a quality-aware streaming (QAS) algorithm to maximize the network utility, which is the difference of time-averaged users' perceived video quality and time-averaged HWN's transmit power. Simulation results exhibit that the proposed QAS algorithm can significantly improve network utility compared with the state-of-art baselines, which are not specific for on-demand video streaming.

The development of Network Function Virtualization (NFV) and Software Defined Networks (SDN) standards is an opportunity for Mobile Network Operators (MNOs) to deploy Content Delivery Network (CDN) functionalities into the mobile network edge, such as Base Stations (BSs). In this paper, we investigated the content placement problem for the BS-based wireless CDN system. We call the storage resources implemented on BSs as storage helpers. Due to the limited helper storage capacity and the limited user population served per BS, helpers exhibit low hit ratio comparing to traditional CDN edge servers serving a wide area. Then, cooperation is a suitable means to enhance the performance of the wireless CDN system. We propose that BSs close to each other cooperate in replicating content and replying user requests. We formulate the optimum content placement problem to minimize the traffic pressure on mobile network gateways, and show the problem complexity is NP-Hard. We then transform the problem into a multiple-Maximum Weighted Independent Set problem, and propose a heuristic algorithm. The evaluation shows that the hit ratio is improved by our algorithm comparing to the traditional Least Frequently Used (LFU) policy without cooperation.

This work is motivated by the following observation: modern wireless applications such as content prefetching, allow different extent of delay tolerance, thus in practice users may have different time‐averaged data rates during a certain period of time. In this study, the authors consider resource allocation in small cell networks under time‐varying channels while each user has an individual time‐averaged rate requirement. Stochastic optimisation model is employed to minimise the time‐averaged power consumption of small base stations subject to individual user's time‐averaged rate constraint. They develop an online power optimal resource allocation (PORA) algorithm to achieve the optimal power allocation and subcarrier assignment decisions without prior knowledge of channel statistics. Furthermore, considering that the power allocation and subcarrier assignment problem is a nonconvex combinatorial problem, they further develop an iterative heuristic algorithm with polynomial complexity. Simulation results show the effectiveness of PORA and verify the theoretical analysis on the network performance.

The delivery of live video channels for services such as twitch.tv leverages the so-called Telco-CDN-Content Delivery Network (CDN) deployed within the Internet Service Provider (ISP) domain. A Telco-CDN can be regarded as an intra-domain overlay network with tight resources and critical deployment constraints. This paper addresses two problems in this context: (1) the construction of the overlays used to deliver the video channels from the entrypoints of the Telco-CDN to the appropriate edge servers; and (2) the allocation of the required resources to these overlays. Since bandwidth is critical for entrypoints and edge servers, our ultimate goal is to deliver as many video channels as possible while minimizing the total bandwidth consumption. To achieve this goal, we propose two approaches: a two-step optimization where the optimal overlays are firstly computed, then an optimal resource allocation based on these pre-computed overlays is performed; and a joint optimization where both optimization problems are simultaneously solved. We also devise fast heuristic algorithms for each of these approaches. The conducted evaluations of these two approaches and algorithms provide useful insights into the management of critical Telco-CDN infrastructures.

The growth of Internet video traffic imposes a severe capacity problem in today's Content Delivery Network (CDN). Rate-adaptive streaming technologies, such as the Dynamic Adaptive Streaming over HTTP (DASH) standard, reinforces this problem in the core CDN infrastructure since delivering one video means delivering multiple representations for an aggregated bit-rate that is commonly over 10 Mbps. In this paper, we explore better trade-offs between CDN infrastructure cost and Quality of Experience (QoE) of the end-users for live broadcast video streaming applications. We consider in particular underprovisioned CDN networks, our goal being to maximize the QoE for the population of heterogeneous end-users despite the lack of resources in the intermediate CDN equipments. We show that previous theoretical models based on elastic bit-rates do not fit for this context. We propose a user-centric discretized streaming model where the satisfaction of end-users is related to the context and where a stream has to be either delivered in its entirety, or not delivered at all. We first formulate an Integer Linear Program (ILP) that achieves the optimal delivery through a multi-tree delivery overlay. The evaluation of the ILP shows the benefits of this model. We then design a practical system by revisiting the three main algorithms implemented in CDN: user-to-server assignment, content placement and content delivery. At last, we use a realistic trace-driven large-scale simulator to study the performances of our system. In particular, we show that the population of users is reasonably well served (three quarters of the population do not experience degradation) even when the CDN infrastructure experiences a severe underprovisioning (less than half of the required infrastructure).

A Telco-CDN can be regarded as an intra-domain overlay network with tight resources and critical deployment constraints. This paper addresses two problems in this context: (1) the construction of the overlays used to deliver the video channels from the entrypoints of the Telco-CDN to the appropriate edge servers; and (2) the allocation of the required resources to these overlays. Our ultimate goal is to maximize the number of delivered channels while preserving network resources. Two classes of heuristic algorithms, namely two-step optimization and joint-optimization, are proposed to solve these problems. The conducted evaluations confirm the efficiency of the joint-optimization approach.

Content Delivery Networks (CDNs) are confronted with a sharp increase in traffic related to live video (channel) streaming. Previous theoretical models that deal with streaming capacity problems do not capture the emerging reality faced by today's CDNs. In particular, a modern CDN has to deliver a large set of independent non-divisible data streams, which need to be either delivered in whole, or not delivered at all. This constraint is not addressed in previous works. In this paper we identify a new, discretized streaming model for live video delivery in CDNs. For this model we formulate a general optimization problem and show that it is NP-complete. Then we study a practical scenario that occurs in real CDNs. We present a fast, easy to implement, and near-optimal algorithm with performance approximation ratios that are negligible for large network. To our knowledge, these are the first results for the discretized streaming model, and have both practical and theoretical importance in a topic of growing criticality.

We propose a peer-to-peer system for streaming user-generated live video. Peers are arranged in levels so that video is delivered at about the same time to all peers in the same level, and peers in a higher level watch the video before those in a lower level. We encode the video bit stream with rate less codes and use trees to transmit the encoded symbols. Trees are constructed to minimize the transmission rate for the source while maximizing the number of served peers and guaranteeing on-time delivery and reliability at the peers. We formulate this objective as a height bounded spanning forest problem with nodal capacity constraint and compute a solution using a heuristic polynomial-time algorithm. We conduct ns-2 simulations to study the trade-off between used bandwidth and video quality for various packet loss rates and link latencies.