Adlen Ksentini's research while affiliated with EURECOM and other places

... This framework integrates workload prediction and resource pre-provisioning to provide high utilization of edge resources. Pramanik et al. characterize the computational and memory requirements of virtual RANs with regression models to predict better demands for resources [6]. They use 4G vRAN test-bed leveraging the non-disaggregated open-source mobile communication platform and general-purpose processor-based servers. ...

... The mode selection is determined by the method, which is based on the immediate reward provided by the surroundings. In the paper [18], the authors discuss the deep learning-based work that has been proposed to improve the 5G RAN portion of the network, as well as how these works can be integrated with the AI-enabled Open-RAN architecture. ...

... The primary goal of 5th Generation networks is to combine multiple services into a single infrastructure, each with its own logical network [7]. The 5th Generation of mobile networks is the next step in the evolution of mobile networks and is regarded as the digital world's future.5th ...

... • Artificial intelligence (AI) jobs, which usually have a large number of concurrent tasks. Examples include smart home applications [42,82], smart healthcare [46], anomaly detection [81], object recognition [37], autonomous driving services [8], and intelligent photo management [68]. In particular, machine learning jobs that are submitted in TensorFlow [1] framework are a representative type of AI jobs. ...

... IoV uses Vehicle to Vehicle (V2V) and Vehicle to Infrastructure (V2I) for communication [3]. The interactive information mainly includes road congestion, moving tracks, traffic accidents and entertainment consultation to create a safe and comfortable driving environment [4]. In order to prevent criminals from eavesdropping on and tampering with information, which may endanger the traffic system, information interaction must ensure the integrity and security of data, otherwise the system is vulnerable to malicious attacks, including replay, camouflage, message tampering attacks, etc. [5]. ...

... The main aspect of open-source-MEC is the service-oriented MEC layer (Brik et al., 2020), which contains a uniform south-bound functionality interface and varied network functions, with the south-bound functionality interface connecting different network functions jointly based on the uniform stateless hypertext transmission protocol (HTTP) to ensure that they may interact directly with one another if needed. The work also presented the detached centralized service functionalities into distinct (Huang et al., 2017(Huang et al., , 2022aLv et al., 2018). After that, to improve open-source MEC, the work grabbed numerous more network functions from the 5G core network control plane and constructed several new network functions. ...

... The idea of network slicing may be seen as an alternative to network sharing since it creates an end-to-end logical network that utilizes shared infrastructure. According to the network slicing idea, RAN, transport, and CN resources for the network provider are virtualized, and all these resources may subsequently be dynamically or permanently given to various MNOs (Ksentini and Nikaein, 2017;Afolabi et al., 2018). ...

... Therefore, the HBA-NS approach will be a suitable method for network slice placement instead of OCNS and MILP (Papagianni et al., 2018) when resizing network slices periodically to cope with the dynamic nature of the network. Given that UE requests are predicted for each network slice at 10-min intervals in the proposed CNN-LSTM architecture and assuming the startup time for a VNF instance that is hosted in a container is equal to 1 second approximately (Frangoudis et al., 2017), HBA-NS provides the capability to manage network slices in real-time and scaling them dynamically during each prediction interval as HBA-NS completes the network slicing process before the time interval (10 minutes) expires. Fig. 16 illustrates the overall cost of network slice deployment for each approach. ...

... Existing studies on the allocation of computing and communication resources to different network slices in 5G networks can be broadly classified based on their architectures as either one-tier schemes [15][16][17][18][19], two-tier schemes [20][21][22][23][24], as well as Graph-based deep learning and reinforcement learning schemes [25][26][27][28][29]. The related studies are reviewed in the following sub-sections. ...