Sammendrag
In today’s heterogeneous network environments, different applications have varying network resource requirements, such as throughput, processing time, and storage. Traditional approaches to addressing this heterogeneity involve adding additional infrastructure to provide separate resources for each application, ensuring QoS but increasing installation and maintenance costs. Moreover, scalability becomes a challenge as the number of users and emerging applications grow, requiring constant reconfiguration and infrastructure expansion. To tackle these issues and reduce costs while accommodating heterogeneous services within a shared physical network infrastructure, network slicing has emerged as a promising solution. Network slicing allocates physical resources to specific services or groups of services, ensuring guaranteed network resources even during peak periods. However, implementing network slicing comes with implementation complexities and trade-offs in terms of cost, deployment complexity, and system performance.
This thesis aims to investigate a network slicing technique that shares network resources among slices to provide QoS for each service. Various methods for implementing network slicing exist, including client service
division into slices in software-defined network platforms and hierarchical scheduling in hierarchical QoS (HQoS) approaches. We focus on utilizing the DPDK QoS framework for network slicing. We employ a hierarchical token bucket queue and define rules to identify and assign traffic to specific slices. Our approach leverages the DPDK QoS framework in a testbed where a Linux server hosts heterogeneous services in Docker containers, accessed by numerous clients. We fine-tune the control knobs of the DPDK QoS framework and monitor traffic patterns to evaluate its characteristics and efficiency in utilizing shared physical network resources in a heterogeneous applications environment. The contributions of this thesis include the development of a physical testbed for multi-application environments with traffic differentiation, facilitating reproducible and extensible experiments. Additionally, a feasibility study is conducted to assess the effectiveness of the DPDK QoS framework for achieving network slicing and ensuring slice isolation. These contributions advance the understanding and practical implementation of network slicing techniques,
aiding in resource allocation optimization, QoS assurance, and enhanced efficiency of shared physical network infrastructure in heterogeneous multi-application environments.