An SDN Framework for UAV Backbone Network towards

An SDN Framework for UAV Backbone Network towards Knowledge Centric Networking SPEAKER : YU-HSIAN, SU ADVISOR : DR. KAI-WEI KE DATE : 10/20/2021/10/20 1

OUTLINE • INTRODUCTION • SYSTEM MODEL AND PROBLEM DEFINITION • MONITORING PLATFORM • LOAD BALANCING ALGORITHM • SIMULATION AND ANALYSIS • CONCLUSION • REFERENCES 2021/10/20 2

INTRODUCTION • First, massive information exits in UAV networks including the flight and control information of UAVs, the protocol stack information in the UAV network, the sensing information of UAVs, and the obtained information from ground terminals. • Second, the wireless links and network topologies are changing result from the UAV motion. • How to manage and utilize the information, deal with the intermittent links, and at the same time maintain a fluid topology therefore become a challenging problem. 2021/10/20 3

INTRODUCTION(cont’d) • Software Defined Networking (SDN) separates the control plane and the data forwarding plane of the networks, which shows great potential in dealing with unreliable wireless links and provide flexible switching and routing strategies. • SDN provides a solution to programmatically control UAV networks, and therefore makes it easier to deploy and manage new applications and services, especially when the links and topology are changing because of the motion, power failure or replacement of UAVs. 2021/10/20 4

INTRODUCTION(cont’d) • We design a SDN framework considering the dynamic characteristics of UAV networks. • A novel monitoring platform in SDN controller is designed, which can efficiently and flexibly deal with link switching and route selecting. • A load balancing algorithm is proposed to fully utilize the UAV network resources and at the same time avoid traffic congest. • We validate the feasibility and effectiveness of the designed SDN framework in UAV networks with an example that UAVs working as aerial backbone to serve the ground user equipment (UE). 2021/10/20 5

SYSTEM MODEL AND PROBLEM DEFINITION • There are three kinds of communication objects in the considered SDN-based network: UAV, UE and SDN controller 2021/10/20 6

SYSTEM MODEL AND PROBLEM DEFINITION (cont’d) • In the SDN-based UAV network, the controller gathers network statistics and parameters of UAVs, and then utilizes the precise computed results to make the optimal decision. 2021/10/20 7

SYSTEM MODEL AND PROBLEM DEFINITION (cont’d) • Therefore, the core two problems can be summarized as follows: • How to manage the large amount of information in the UAV network including routing protocols, UAV flight control, network programming, security mechanism and so on. • The communication duration and performance of the UAV network are fundamentally constrained by the limited energy storage. • So, it is important to fully utilize the UAV resources and prolong the life time of whole the UAV network. • This calls for more effective resource management and optimization decision (link switching and routing) specifically designed for UAV communication systems. 2021/10/20 8

MONITORING PLATFORM • We propose a monitoring platform in the SDN controller to obtain and process the network knowledge, and its architecture is shown in Fig. 3. 2021/10/20 9

MONITORING PLATFORM (cont’d) • A dictionary for each flow is given by • where AB indicates a link between UAV A and B, and the same for other UAVs. • We add each flow layer which is the dictionary defined in (1) to get the network topology and link load as shown in Fig. 4. The dictionary for network topology is given by 2021/10/20 10

MONITORING PLATFORM (cont’d) • Therefore, the network topology and load for each link can be acquired from the dictionary as shown in. 2021/10/20 11

MONITORING PLATFORM (cont’d) 2021/10/20 12

LOAD BALANCING ALGORITHM • There are three main aspects considered in our algorithm, which are described as follows. • First, global variance and local variance are considered to determine whethere is load imbalance in the network. • The global variance is the network variance and local variance is the flow load. • Second, we distinguish each flow with different priorities, which is necessary to provide better quality of service (Qo. S) to the UEs with higher priorities. • Third, we consider the power limit of UAVs, since when calculating the optimal path for flow we utilize the UAV battery information. 2021/10/20 13

LOAD BALANCING ALGORITHM (cont’d) • Fig. 6 shows the flowcharts of the load balancing algorithm in the SDN controller. 2021/10/20 14

LOAD BALANCING ALGORITHM (cont’d) • The controller monitors network and evaluates the link state by the network variance and flow load. • The flow load indicates the total load of links that one flow passes through. And the network variance is given by 2021/10/20 15

LOAD BALANCING ALGORITHM (cont’d) • We utilize Dijkstra algorithm with the network topology to calculate the optimal path and the weight is the load for each link. • When the optimal path is found, the UAV’s battery storage is considered. • If the UAV’s power consumption exceeds the power threshold which indicates that it is out of power, the UAV and the links it connects to are removed from the topology and we replan the routing path in the new topology. • When the optimal path is determined, the device with the highest priority is switched first according to the priority of the devices. The link load state will be detected again after a switchover. 2021/10/20 16

SIMULATION AND ANALYSIS • In this section, we implement the proposed monitoring platform and the load balancing algorithm into Mininet-Wi. Fi and evaluate them with computer simulation. We adopt the POX controller as the SDN controller. Other primary simulation parameters are listed in Table I. 2021/10/20 17

SIMULATION AND ANALYSIS (cont’d) • 16 UEs (denoted by Sta x) are deployed on the ground while 5 UAVs (AP) form a mesh network to provide communication service in the sky which is shown in Fig. 7. • Sta 1~4, 6, 7, 10, 16 are set as clients which send UDP traffic at a data rate of 1 Mbps to the servers which are Sta 11~15, 17, 19, 20, and the flow directions are as shown in Fig. 7. • The paths between Sta 1~4 and Sta 11~14 are all AP 1 -AP 5 -AP 4. The path between Sta 16 and Sta 20 is AP 4 -AP 3 -AP 2. • The rest of UE pairs are connected with each other through direct UAV paths, for example, the flow from Sta 6 to Sta 9 can be directly forwarded by AP 1 to AP 2. 2021/10/20 18 • We use i. Perf tool to generate traffic in the UAV network.

SIMULATION AND ANALYSIS (cont’d) 2021/10/20 19

SIMULATION AND ANALYSIS (cont’d) • A. Load balancing without battery consideration • In this subsection, we evaluate the load balancing algorithm without considering the battery limit, which means that the flow from overloaded APs can be transferred to any other APs. 2021/10/20 20

SIMULATION AND ANALYSIS (cont’d) • When all UEs start to work, the SDN controller gathers all the network statistics and calculates the network topology. • There are 4 UEs sharing the link between AP 1 and AP 5, and 5 UEs sharing the link between AP 5 and AP 4. • As a result, the monitor detects that the flow between AP 1 and AP 4 is overload and the network variance is large because of the unbalanced link load distribution. • It collects the device information of Sta 1~4 and utilizes Dijkstra algorithm to get the optimal path without considering the power limit of UAVs. 2021/10/20 21

SIMULATION AND ANALYSIS (cont’d) • The number of packets processed in APs in one period is shown in Fig. 8. The blue bar indicates the number of packets in APs without load balancing algorithm while the red bar indicates the number of packets in APs with our load balancing algorithm. 2021/10/20 22

SIMULATION AND ANALYSIS (cont’d) • To provide a deeper look, the detailed load variations of AP 2, AP 3 and AP 5 are shown in Fig. 9. It is clear that the traffics in AP 2, 3 and 5 are changing with the increasing period until the load balancing is achieved (at period 20). 2021/10/20 23

SIMULATION AND ANALYSIS (cont’d) • The first flow (red dotted line) is transmitted from AP 5 to AP 2 when the flow load and the network variance exceed their threshold, i. e. , 5 and 1 respectively as in Table I, for 10 consecutive time periods (the first 10 periods), which means that the AP 5 is overloaded. • And then, Sta 1 with the highest priority changes the path from AP 1 AP 5 -AP 4 to AP 1 -AP 2 - AP 4, similarly, Sta 2 also changes the path from AP 1 -AP 5 - AP 4 to AP 1 -AP 3 -AP 4 in the second 10 periods. • In the second 10 periods, AP 5 is still overloaded, so its flow is further transmitted to AP 3. • Gradually, the numbers of packets in AP 2, AP 3, AP 5 are getting closer which indicates that our load balancing algorithm works well in the 2021/10/20 24 UAV network.

SIMULATION AND ANALYSIS (cont’d) • B. Load balancing with battery consideration • As we know UAV network is an energy constrained network. Therefore, we take the power limit of UAVs into consideration. • We assume that AP 3 is with low battery which means that it is impractical to increase its traffic. • The electricity consumption of AP 3 is set to 80% in the simulation which means that the power threshold is exceeded. 2021/10/20 25

SIMULATION AND ANALYSIS (cont’d) • As shown in Fig. 10, the number of packets in AP 3 is unchanged after applying the load balancing algorithm which indicates that the traffic is all transmitted to AP 2 instead of AP 3. • Fig. 11 shows that no extra traffic flows into AP 3 and the load balancing works between AP 5 and AP 2. • SDN controller take the UAV battery information into consideration in finding the optimal path. • Therefore, the optimal path for Sta 2 is AP 1 -AP 2 -AP 4, corresponding to the load increase of AP 2 as shown in Fig. 11. 2021/10/20 26

SIMULATION AND ANALYSIS (cont’d) 2021/10/20 27

SIMULATION AND ANALYSIS (cont’d) • In Table II, we list the network variance to show the effect of network load balancing algorithm. • For all the links in the network, smaller difference in the traffic load between the links indicates the better load balancing effect of the network. • We can see that each time after switching the network variance decreases, and it validates that the proposed load balancing algorithm performs well. 2021/10/20 28

CONCLUSION • A monitoring platform in SDN controller and a load balancing algorithm are proposed. • The proposed load balancing algorithm which utilized the analysis results from the platform is also validated to be effective in the traffic balancing. • We also consider the power limit of UAVs and the results indicate that our algorithm performs well in energy constrained network. • We believe that the reported framework in this paper is scalable to adopt machine learning to analyze network datasets and therefore may contribute to the future KCN. 2021/10/20 29

REFERENCES • Zhang, Xiao, et al. “An SDN Framework for UAV Backbone Network towards Knowledge Centric Networking. ” IEEE INFOCOM 2018 - IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), 2018, doi: 10. 1109/infcomw. 2018. 8406959 2021/10/20 30