Emerging latency-critical applications such as cloud gaming and industrial automation demand agile and ultra-low-latency service delivery, which traditional network appliances struggle to support. Network Function Virtualization (NFV) addresses this by chaining Virtual Network Functions (VNFs) into Service Function Chains (SFCs). Parallelized SFCs (PSFCs) reduce service delay by executing independent VNFs concurrently, but introduce significant copy/merge and buffering overheads due to synchronization delays across branches. Moreover, dynamic PSFC arrival rates complicate efficient VNF placement decisions. This paper presents RL-PARETO, a hybrid deep reinforcement learning approach that adaptively orchestrates parallel VNFs while minimizing parallelization overheads and satisfying SLA constraints. RL-PARETO uses a graph transformer encoder with dual pointer-network heads to jointly generate PSFC partitions and VNF placements in a single pass. Training integrates Proximal Policy Optimization (PPO) for stable exploration with a Double-DQN critic for efficient value estimation. A fallback heuristic ensures feasible deployments under resource constraints. Extensive evaluations across diverse network topologies demonstrate that RL-PARETO achieves up to $10 \%$ higher acceptance rate and $15 \%$ reduction in merge buffer overhead, while maintaining robust performance under dynamic conditions.

Next-Generation Radio Access Networks (NG-RANs) aim to facilitate high data rates, low-latency applications, and dense mobile connectivity — benefit from the integration of Artificial Intelligence and Machine Learning (AI/ML) to enhance performance and efficiency. Nevertheless, the dynamic service demands within NG-RAN (namely Open RAN) lead to AI/ML performance degradation known as drift, resulting in violations of Service Level Agreements (SLA) and issues like over-or under-provisioning of resources. Detecting and adapting to drift becomes crucial to meet the diverse requirements of intelligent networks. Due to frequent retraining, the existing threshold and classifier-based approaches have potential disadvantages such as SLA violations and resource inefficiency. This paper introduces a novel approach that exploits the Generative Adversarial Network (GAN) architecture to determine the drift and anomaly. The proposed approach is evaluated for a throughput prediction use case over a real-time dataset and compared to the threshold and classifier-based approaches. The results show that the proposed approach outperforms the threshold and classifier-based approaches.

Network Functions Virtualization (NFV) offers enhanced programmability and cost-efficiency by replacing hardware middleboxes with versatile Virtual Network Functions (VNFs) on commodity servers. Key challenges in NFV orchestration involve economically placing VNFs and chaining them to meet the requirements of the Service Function Chains (SFCs). Additionally, depending on the VNF functionality, some VNFs may be shared or may not be shared among SFCs. Many existing studies have overlooked the potential benefits of placing shareable VNFs on high centrality nodes, which can boost re-usability and increase SFC acceptance ratios. Vertical scaling of VNF resources, rather than over-allocating for shareable VNF instances, can address resource under-utilization for future SFC admissions. This paper introduces a mechanism for SFC placement, considering both VNF reuse and vertical scalability when deploying SFC requests in the network. The proposed mechanism demonstrates significant improvements in SFC acceptance ratios compared to two baseline schemes, SPH-BC and SPH-VS, as well as a state-of-the-art scheme, SMA-VAA.

Enabling widespread adoption of resource-intensive vehicular applications such as Extended Reality (XR) and High Definition map (HD Map) necessitates further enhancements in 5G, which is anticipated with 5G-Advanced. These applications, sensitive to latency, prompt researchers to propose offloading vehicles' complex computations to nearby edge clouds, aiming to minimize latency and meeting the Quality-of-Service (QoS) demands of these applications. However, the uncertainties arising from spatio-temporal factors due to vehicle mobility and the dynamic nature of application behaviour pose significant challenges in deciding the efficient offloading decision for minimizing latency. To tackle this challenge, this paper introduces a crosslayer framework that bridges the Radio Access Network (RAN) scheduler with the Mobile Edge Computing (MEC) scheduler. The proposed framework facilitates the exchange of vehicle ranks and channel condition information between schedulers, strategically aimed at reducing Head-Of-Line (HOL) delay for efficient computational offloading. Furthermore, the MAC layer incorporates the prediction of the Buffer Status Report (BSR) using Machine Learning (ML) to further reduce the queuing delay experienced by the offloading jobs of the vehicles in uplink. Simulation results using the NS-3 gym demonstrate that the proposed cross-layer framework achieves a higher Offloading Success Rate (OSR) than the state-of-the-art QoS scheduler by effectively reducing HOL delay for HD Map vehicular application.

Beyond fifth-generation (B5G) networks (namely 6G) aim to support high data rates, low-latency applications, and massive machine communications. Integrating Artificial Intelligence (AI) and Machine Learning (ML) models are essential for addressing the network’s increasing complexity and dynamic nature. However, dynamic service demands of B5G cause the AI/ML models performance degradation, resulting in violations of Service Level Agreements (SLA), over-or under-provisioning of resources, etc. To address the performance degradation of the AI/ML models, retraining is essential. Existing threshold and periodic retraining approaches have potential disadvantages such as SLA violations and inefficient resource utilization for setting a threshold parameter in a dynamic environment. This paper presents a novel algorithm that predicts when to retrain AI/ML models using an unsupervised classifier. The proposed predictive approach is evaluated for a Quality of Service (QoS) prediction use case on the Open RAN Software Community (OSC) platform and compared to the threshold approach. The results show that the proposed predictive approach outperforms the threshold approach.

Network Functions Virtualization (NFV) enables carriers to replace dedicated middleboxes with Virtual Network Functions (VNFs) consolidated on a few shared servers. However, the question of how (and even whether) one can achieve performance related Service Level Objectives (SLOs) with software packet processing in NFV remains open. VNF consolidation causes high variability and unpredictability in throughput and latency of VNFs deployed together. It was shown in our prior work that isolating the processor’s Last Level Cache (LLC) and limiting Memory Bandwidth (MB) directly helps in achieving performance isolation among the co-located VNFs. So, in this work, we formulate VNF placement problem with exclusive allocation of LLC and MB resources as a Mixed Integer Linear Program (MILP). Due to its hardness to solve, we also present a heuristic solution named RAVIN that enforces performance SLOs for multi-tenant NFV servers while being as much resource-efficient as possible. We demonstrate RAVIN’s effectiveness in improving resource utilization and in reducing the total number of required servers to deploy VNFs compared to state-of-the-art and baseline approaches.

Mobile operators are widely adopting Network Functions Virtualization (NFV) to get the benefits of virtualization, including ease of deployment, flexibility, and cost savings. NFV allows multiple virtualized Radio Access Networks (vRANs) to run on commodity hardware enabling joint signal processing and efficient interference management. In addition, mobile operators can run general-purpose workloads alongside vRANs to utilize spare system resources in the NFV infrastructure. In such a consolidated scenario, it is necessary to ensure that the Key Performance Indicators (KPIs) of vRAN workloads are always met. But the workload consolidation could cause high variability and unpredictability in the performance of the deployed vRANs due to contentions for shared system resources like CPU cores, Last Level Cache (LLC), etc. In order to address this problem, we present JARS – a joint allocation of radio and system resources for the NFV infrastructure – that dynamically adjusts system resources such as CPU cores and LLC-ways, and radio resources such as Physical Resource Blocks (PRBs) to ensure KPIs for vRANs and improve overall resource utilization by workload consolidation. We profile srsLTE to determine the minimal CPU core and LLC resource requirements to satisfy KPIs during different traffic loads, which is used in the JARS. Experimental studies on a prototype system show that the proposed JARS outperforms a state-of-the-art scheme by 23%.

Vehicular applications such as Augmented Reality (AR), Virtual Reality (VR), and High Definition Map (HD Map) are known for their latency-sensitive traits. But, dynamic scheduling at the MAC layer incurs significant signalling overhead (in terms of Scheduling Requests (SRs) in Uplink (UL)), leading to non-negligible latency in 5G NR. To address this issue, 5G NR introduces Configuration Grant (CG) for UL transmission, which pre-allocates radio resources to UEs (vehicles), thereby reducing signalling overhead between a vehicle and the Base Station (gNB). However, the high-speed mobility of vehicles results in rapid changes in channel conditions. Employing CG in a vehicular scenario can lead to incorrect assignment of transmission parameters (e.g., Modulation and Coding Scheme (MCS)), thereby adversely impacting the vehicles’ Packet Delivery Ratio (PDR). To address this issue, this paper proposes a CG allocation algorithm that utilizes a Machine Learning (ML)-driven approach to predict the future MCS of vehicles. A data-driven ML model, derived from a real-world dataset, assists the radio resource scheduler and is evaluated using the NS-3 5G-LENA CG module. The ML-assisted CG allocation algorithm demonstrates significant improvements in terms of PDR and spectrum usage efficiency in vehicular scenarios.

Network Functions Virtualization (NFV) promises programmability and cost savings by replacing hardware middleboxes with more flexible Virtual Network Functions (VNFs) on commodity servers. But, the current virtualization technologies do not fully isolate the system resources like Last Level Cache (LLC) and Memory Bandwidth (MB); therefore, co-location of VNFs on the same commodity server causes interference effects which might severely impact the performance of VNFs in terms of throughput, latency, etc. Contention at LLC is one of the root causes of this performance degradation and it is addressed by LLC resource partitioning. But, it remains unexplored the impact of both LLC and MB on VNF performance. In this work, we investigate the importance of MB partitioning along with LLC partitioning to achieve performance isolation in NFV-based systems. Allocating these system resources among co-located VNFs to meet Service Level Agreements (SLAs) is challenging due to the dynamic nature of traffic and varying functionality of VNFs. In this work, we formulate the resource allocation problem as an Integer Linear Programming (ILP) for maximizing the number of accepted VNF requests with SLA guarantees. Since the problem is NP-hard, we present a polynomial time$\epsilon $-approximation scheme. Further, we propose a heuristic approach namedNFVPermit, a resource manager for NFV-based systems that tries to ensure performance isolation among co-located VNFs based on their current traffic rates and SLA requirements. Through extensive experiments, we show howNFVPermitoutperforms state-of-the-art and baseline approaches.

To reduce the processing delay from the sequentially running virtual network functions (VNFs) in a service function chain (SFC), network function parallelism (NFP) is introduced that allows VNFs of the SFC to run in parallel. Existing NFP solutions only focused on improving parallelism benefits without paying much attention to resource utilization while deploying VNFs of SFCs. We take advantage of resource-delay dependency to propose a flexible and efficient parallelized SFC placement mechanism called FlexSFC which determines the optimal SFC placement while reducing resource usage and meeting end-to-end delay guarantees of the SFCs deployed. Initial results show that FlexSFC guarantees the end-to-end delay requirement with better resource utilization and SFC acceptance rate than the state-of-the-art approaches.

The advent of Open Radio Access Network (O-RAN) technology enables intelligent edge solutions for base stations in beyond 5G (B5G) networks. O-RAN Working Group 2 (WG2) focuses on the architecture and specifications of AI/ML workflows, allowing AI/ML applications in O-RAN environments to meet different QoS requirements for different use cases over varying time periods. This study shows the technical challenges in mapping AI/ML functionalities at Near-Real Time (RT) RAN Intelligence Controller (RIC) and/or Non-RT RIC for closed loop control-based resource adaptation in O-RAN. We also present a drift-based solution to avoid performance violations if there is decay in prediction accuracy. Results show that drift-based solution outperforms offline models.

Service Function Chaining (SFC) is used to steer the traffic to a specific set of Network Functions (NFs) (such as load balancer, proxy, firewall, etc.) based on the type of traffic and operator policy. Handling the massive amount of user traffic envisioned in the next generation networks using traditional techniques is costly and tedious. By leveraging advanced technologies such as Network Functions Virtualization (NFV) and Software Defined Networking (SDN), NFs can be deployed as software instances on Virtual Machines (VMs) (also called as Virtual Network Function (VNF)). Network operators widely place different types of VNFs at different locations to meet the user traffic demands. Multiple VNF instances on the same physical server compete for common resources such as network I/O bandwidth, CPU cycles, cache memory, and main memory which can lead to severe performance interference, which is ignored in existing NF selection mechanisms. However, increasing the SFC acceptance rate of SFC requests with an effective selection of required VNFs under the constraint of end-to-end latency is still an open problem. Since this problem is NP-Hard, we propose a heuristic algorithm based on dynamic programming which efficiently selects the required VNFs and steers the traffic by considering the interference effect. Results show that the proposed algorithm improves the average SFC acceptance rate by 29% as compared with existing methods.