Моделирање и анализа на перформанси во SDN-базирани мобилни јадрени мрежи

Abstract

The Software-Defined Networking (SDN) is a concept that is already widely used in data centers and mobile core networks to provide network-wide abstraction, open interfaces, programmability, and hiding of network complexities. In today’s Long-Term Evolution (LTE) implementations, there is no guarantee for a given latency in the network. Furthermore, the LTE architecture does not distinguish between services that require a certain level of latency performance from the services that do not have stringent latency requirements. The upcoming 5th generation (5G) of mobile networks is currently still being standardized to accommodate a large number of services and applications which have very diverse requirements. There are already many use cases defined that require very small guaranteed latency in the region of 1 ms to 100 ms. Fault recovery mechanisms in the 5G core network must support and enable the services for reaching the low latency values. In this thesis, an optimized protocol for fast failure detection and recovery that is implemented directly in the data plane and is suitable for the 5G Evolved Packet Core (EPC) is proposed. The novel protocol uses the OpenFlow’s (OF) fast-failure mechanism for local faults, and a more advanced mechanism with stateful user plane recovery protocol that is activated only in the case of remote faults. The simulation results prove that the proposed solution can achieve fast fault recovery times and can be implemented in the future 5G core networks. Furthermore, the major latency contributors in OF-based mobile core and backhauling networks are investigated and mathematically modeled. By using queuing theory, the average packet service time in both switches and controllers are quantified. The results clearly point the major factors that impact the latency in SDN networks: load and number of controllers, arrival rate, mean service rate and probability of Packet-In messages. In this thesis, the aim is to also evaluate the impact of some of the assumptions commonly used when mathematically modelling the performance of SDN-based mobile networks. By applying queuing theory, the impact of the following assumptions is quantified: load independent service rate, and limited buffer size. The results show that considering a limited buffer significantly impacts the accuracy of the modelling, while at load dependent service rate а visible service time deterioration at high loads is noticed. A centralized controller approach in SDN creates problems of network scalability and performance degradation, which is the main reason for introducing an architecture of logically distributed controllers. Ensuring the continuity of a user’s ongoing session and reducing the interruption time when performing a handover, is one of the major challenges in today’s modern mobile networks. In this dissertation, a novel analytical approach to model the delay introduced by the handover-related OpenFlow signaling messages in SDN networks with multiple controllers is proposed. The results reveal that high probability of Packet-in messages causes rapid degradation of the network performance. Similarly, as the number of controllers increases, the number of synchronization messages intensifies, and this impacts negatively the packet service time. The findings can be used when designing handover delay guarantees in a mobile network with a target system throughput. Finally, two system models with specific OF-switch design are analyzed: 1) single buffer with no prioritization for both control and user plane; 2) two isolated buffers with different priority for handling the control and user plane separately. The two systems are compared by analyzing the total handover delay and the needed buffer size. The conclusion is that using priority buffering in the switch should be the preferred design choice for mobile networks, as it provides smallest handover delay and has the smallest buffer needs in the scenarios of interest.