Resumen de Enhanced quality of service mechanisms for 5g networks
The heterogeneous services with stringent requirements that are envisioned to co-exist under the fifth generation of cellular networks (5G), inexorably challenge the current Long-Term Evolution (LTE) cellular network’s features. Consequently, an important amount of effort has been invested to reduce current latency while increasing the throughput and the reliability. As an example, 5G’s uplink granted free transmission procedure permits reserving Resource Blocks(RBs) in advance for a set of User Equipments (UEs), and thus, reduce the latency by eliminating the UEs resource request procedure. Likewise, the cellular network stack has experience major changes in different layers. Examples of it is the new Radio Link Controller (RLC) Packet Data Unit (PDU) that enables a faster packet forming at the expense of reducing the compression ratio. These solutions address latency causes that lie in 5G’s protocol stack, and will unquestionably enhance 5G’s capability to meet the rigorous services’ latency requirements. In addition, 5G has introduced a new sublayer (i.e., Service Data Adaptation Protocol (SDAP)), and has defined a new Quality of Service (QoS) Flow Indicator(QFI) as its finest quality granularity indicator. However, delays generated by the pace at which data packets are forwarded or its sizes, can significantly impact the latency in contemporary cellular networks, and ruin the stringent timing guarantees required by delay-sensitive services if they are not carefully considered. One of the problems associated with the pace at which the data packets are forwarded is the bufferbloat, and will specifically occur in 5G’s Radio Access Network RAN since contemporary wired links are orders of magnitude faster than wireless links, and are provided with large buffers to always fulfill the fluctuating radio link capacity.
Being the RAN the bottleneck and having large buffers together with the fact that most of contemporary data is transported through the loss-based congestion control algorithm, Transmission Control Protocol (TCP) Cubic, suffice to bloat the buffers and increase the latency. In this thesis we provide quantitative results of the bufferbloat problem in contemporary cellular networks. We first thoroughly explain 5G’s QoS hierarchical multi-queuing and show how the bufferbloat increases the latency. Following, we propose the (enhanced)5G Bandwidth Delay Product ((e)5G-BDP), the Dynamic RLC Queue Limit (DRQL) and the UPF-SDAP Pacer (USP) solutions, enhancing the current 5G QoS multi-queuing architecture, and approaching through two different paradigms. We evaluate our proposed solutions in an emulator, as well as in a testbed, against state-of-the-art solutions and conclude that a new QoS hierarchical multi-queuing architecture is needed in 5G to fulfill the latency requirements for which it is envisioned. Moreover, since information in the wired link is transported in packets, while in the RAN is transmitted through RBs, packets that do not fit in the assigned RBs, are segmented and transmitted during different Transmission Time Intervals (TTIs). Such mechanism prevents wasting the scarce wireless transmission opportunities. However, the segmented information at the receiver cannot be forwarded until all the remaining information from the packet is reassembled, which in the best case occurs during the next transmission opportunity. We exhaustively study the problem and propose a RB scheduling algorithm named Elastic Quantum Partition (EQP) to address this challenge and compare it quantitatively against a Fixed Partition (FP) RB distribution in different scenarios in a testbed with dynamic Modulation and Coding Scheme (MCS), off-the-shelf equipment and slices. The outcome shows a latency reduction when scheduling the RBs elastically rather than using a fixed scheduler. In summary, in this thesis we shed some light in the bufferbloat phenomenon and the segmentation/reassembly procedure in current cellular networks.
