R + D
Traffic prediction Virtualization in 5G networks allows the realization of network slicing, which is a technology that allows for virtual networks (referred to as Network Slices) to be created on the same physical network. Since the amount of network slices deployed for execution is expected to be very large, it is necessary to automate the […]
Virtualization in 5G networks allows the realization of network slicing, which is a technology that allows for virtual networks (referred to as Network Slices) to be created on the same physical network. Since the amount of network slices deployed for execution is expected to be very large, it is necessary to automate the management of the resources to control resource access by the network slices. For this, multiple AI solutions will be implemented as 5G technology moves into its deployment stage. At Iquadrat, we are targeting this resource management problem by temporally predicting the load that each network slice is expected to handle. Initially, we have started by designing different Deep Learning agents with innovative learning procedures to optimize them for traffic prediction at the base station levels. Using this approach, we will then design full mechanisms (algorithms and implementation) to manage and orchestrate RAN resources. At future stages, we will exploit the prediction capabilities of our deep learning agents to predict resource demand at different points of the 5G network and propose a complete mechanism for resource demand prediction and orchestration at every technological domain. In this way, we will be able to develop complete solutions for Network Slice Admission control, dynamically and statically.
Modern 5G networks consist of an excessive amount of distributed Virtual Machines (VMs) that are working in harmony to provide a single service. Manually managing these VMs to provide maximum efficiency at minimum cost is almost impossible. We have developed an application that is capable to perform zero-touch, intra-slice VM re-configuration, and resource allocation with no human intervention. Our algorithm is based on state-of-the-art, multi-agent deep reinforcement learning and can reconfigure the VM location and chaining dynamically throughout the network. The re-configuration is based on collected data from distributed agents in the network such as the capacity and load of connecting links, the congestion level of local and alternative computing resources, or past measurements. The agents are running locally in different domains of the network, such as the Edge or Cloud domains, and controlling subsets of computing resources, links, and VMs. Their task is to generate local re-configurations that affect only a part of the data flow between VMs, without the need to re-configure the VMs in the entire network. The performance of each slice is constantly measured and evaluated by the agents. A confidence metric quantifies the confidence in the ability of the agents to resolve any potential Service Level Agreement violation or performance issue. If the confidence is below a threshold, then the agents will initiate a wider reconfiguration. This approach offers a dynamic, distributed, elastic, and cost-effective way to orchestrate VMs and their data flow, without the need for expensive global analysis and reconfiguration.
For more information on our networking solutions, please contact us and we will reply as soon as possible!
Las cookies necesarias son absolutamente esenciales para que el sitio web funcione correctamente. Esta categoría solo incluye cookies que garantizan funcionalidades básicas y características de seguridad del sitio web. Estas cookies no almacenan ninguna información personal.