Solutions to Support Mobility in SDN-Based Dense Wireless Networks Speaker: Hsu,Hsuan-Yi Advisor: Ke,Kai-Wei Date: 2016/12/28 1

Outline Introduction Network architecture to support mobility in DN networks Simulation Conclusion Reference 2

Introduction 3

Introduction Dense Networks: A dense network is a network in which the number of links of each node is close to the maximal number of nodes. Each node is linked to almost all other nodes. 4

Introduction Features of Wireless Dense Networks: Small inter-cell distance Large amount of handover events Co-channel interference Changing channel capacity 5

Introduction Common ways to solve problems in Dense Wireless Networks: Macro cells: provides radio coverage served by a high power cellular base station (tower). Multicast: Same data is sent to multiple radio nodes and the schedulers of macro cells are tightly coordinated to transmit data to users. Advantages: Decreased handover event Disadvantages: Higher power consumption Complex transmission coordination 6

Introduction Another way to solve problems in Dense Wireless Networks: The software-defined network (SDN) concept is a strong candidate to provide global network management, the traffic engineering (TE) optimization can be dynamically made by central controller. However the application of SDN concept in dense wireless networks is not straightforward. Unlike wired networks, the channel capacity of wireless networks always changes. 7

Network architecture to support mobility in DN networks Transmission issues Traffic Engineering 8

Network architecture But in dense wireless networks, mobility of nodes makes the global traffic optimization difficult and imprecise. 9

Network architecture Control traffic engineering in fixed network Large but fixed Control traffic engineering in wireless network Smaller but keep changing 10

Network architecture Virtual User-Specific Serving Gateway: Reducing the impact of UE mobility to the complexity of network management and resource allocation. If a v-u-SGW is close to the source and far from the UE, network resources may not be best utilized (for example, when traffic inflation occurs between the v-u-SGW and respective UE), and TE complexity is not well controlled. If, on the other hand, it is near the UE and far from the source, it will not mask UE mobility well and will cause frequent v-u-SGW migration (i.e. replacement), unstable TE decision, and therefore increased control overhead. A network node may host multiple v-u-SGWs. The number of selected v-u-SGW hosts should be kept as small as possible for controlling operating cost. A good trade-off needs to be found during v-u-SGW placement. 11

Transmission issues Duplicated packets: Different layers in the network have their own packet protection mechanisms. These mechanisms can be conflicting and bring unnecessary complexity to the system operation. Transport layer Since the capacity of wireless channel in DN can quickly vary because of mobility, the packet delay is also varying. Hence the TCP sender may be unable to reliably estimate the packet delay for automatic packet retransmission based on packet round-trip time(RTT). The TCP source may resend packets to the serving cells and cause packet duplication. Physical layer HARQ retransmissions cause certain additional delay. In long-term evolution (LTE) systems, the minimum delay is 8 ms. The TCP automatic retransmission may be triggered and thus cause packet duplication. 12

Transmission issues 13 Changing transmission rate Increased packet lost rate Increased retransmission Increased packets on the network 13

Traffic Engineering One promising solution to avoid packet duplication is fountain codes (FC) in combination with multipath TE. This protocol is called fountain coded multipath transmission (FC-MP) protocol. Fountain Coding Transmission: k packets of a file can potentially generate limitless encoded packets; once you receive any k(1+α) packets ,regardless of the order (α is a small fraction<1,especially when k is large )you can quickly reconstruct the original file. Advantages: No need for receivers to send any message to sender to request retransmission 14

Traffic Engineering Fountain Coding Transmission: To benchmark the performance of FC-MP protocol, which heavily depends on the optimality of TE, we assume that the backhaul capacity is significantly larger than that of wireless channels. In this case, the v-u-SGW could send coded packets at the same high rate to multiple serving radio nodes. Radio nodes perform independent packet scheduling to transmit coded packets the receiver. However, due to different wireless channel capacity, radio nodes may send coded packets at different rates to a single receiver. We call this protocol is fountain coded multicast (FC-MC) protocol. 15

Traffic Engineering Fountain Coding Transmission: In our solution, the v-u-SGW collects data packets of a flow and use systematic FC to create coded packets. The coded packets are sent over multiple paths to multiple serving radio nodes. Radio nodes perform independent packet scheduling and transmit FC packets to the receiver until the receiver can decode the original data file. 16

Traffic Engineering Packet redundancy issue : Due to the imprecise rate allocation, the incoming rate of traffic flows to radio nodes could be higher than that can be handled by radio nodes. This leads to a waste of backhaul capacity and reduces the overall network performance. To mitigate this issue, we introduce a feedback mechanism so that radio nodes send back the buffer status to the v-u-SGW, where FC packets are generated. Buffer status reports could be sent at a higher frequency (e.g. every 0.1 second) while the TE is run at slower timescale (e.g. 0.5 second).the data rate of individual flow can be adjusted in between two TE runs according to its buffer size at radio nodes 17

Simulation Simulation Settings Simulation Results 18

Simulation Settings 19

Simulation Settings Mobility model: A street-liked mobility scenario is considered, where users travel in parallel paths through the network. User devices monitor the downlink path losses and report the path losses of the best N serving cells. Traffic model: We investigate performance of best-effort (BE) traffic. In BE sessions, users download files of the same size 20 Mbit. The off-time between two sessions follows an exponential distribution with a given mean off-time, which is set to 10 seconds (low intensity) and 1 second (high intensity). The total simulation time is 3 minutes. Performance: The performance of network is measured by the number of completed downloads. 20

Simulation Settings Comparison: TCP-D TCP provides functions for reliable delivery, rate control and congestion avoidance. In SDN networks, the routing and rate allocation are performed by TE. Hence, these two functions of TCP can be eliminated, only the automatic packet retransmission function could be kept. We call this protocol as TCP- Decomposition (TCP-D) in this study. FCMP FCMC 21

Simulation Results Simulation sets: Purpose: To find out network performance between different transmission protocol and different intensity of off-time. Simulation 1 Low intensity Dropping handover TCP-D FCMP FCMC Compared to Single path TCP-D High intensity 22

Simulation Results Simulation result for BE traffic: Low intensity High intensity 1-path TCP-D 347 completed sessions 1-path TCP-D 927 completed sessions The role of multipath TE is more dominant if traffic intensity increases. With low traffic intensity, radio nodes are underutilized most of time. Therefore giving more paths (or serving cells) can provide only small gains. At high traffic intensity, some radio nodes are heavily loaded. Multipath transmission will improve load balancing and definitely bring larger gains over single path. 23

Simulation Results Simulation set: Purpose: To compare the performance between different handover types. Simulation 2 High intensity Forwarding handover TCP-D FCMP Compared to Dropping handover 24

Simulation Results Impacts of packet forwarding handover: Simulation compared to the packet dropping handover the packet forwarding handover and TCP retransmission may results in packet duplication and thus increases the file downloading time The forwarded packets may slightly help to improve the performance of FC-MP protocol as more FC packets are available when the users' channels have higher capacity. 25

Simulation Results Simulation sets: Purpose: To observe the number of unused FC packets during different TE interval. Simulation 3 FC TE every 0.5 seconds TE every 0.1 seconds TE every 0.5 seconds and feedback to v-u-SGW every 0.1 seconds 26

Simulation Results Reduced number of unused FC coded packets: Since the TE decision is not well-matched with the varying channel capacity, and also due to mobility handover, a number of FC packets may be overly generated. It is observed that by reducing time between TE runs from every 0.5 second to every 0.1 second, the number of unused packets can significantly decreased at the cost of computational. 27 Cumulative density function of number of unused FC coded packets, 2pathTE,100 km/h.

Simulation Results Simulation sets: Purpose: To observe the performance between different cell size. Simulation 4 FC in DN (57 radio nodes in an area of 0.04 𝑘𝑚 2 ) Compared to Macro-cell network (57 radio nodes in an area of 1 𝑘𝑚 2 ) 28

Simulation Results DN vs. Macro-cell Network: The relative gain of DN over macro-cell network is significant for any mobile speed. When the user density increases, the gain also increases. Relative gain of dense small networks compared to macro-cell network. 29

Conclusion 30

Conclusion In this paper, we introduced integrated solutions combining traffic engineering and fountain codes to support mobility in dense wireless networks. We showed that TCP-liked protocol is inefficient in SDN -based dense wireless networks with high mobility users. In the future, our work can be extended in some directions to support high mobility users, for example to further reduce the number of unused (overly generated) fountain coded packets. 31

Reference 1. Solutions to Support Mobility in SDN-Based Dense Wireless Networks Ngoc-Dung Dao; Hang Zhang; Hamid Farmanbar; Xu Li 2015 IEEE Globecom Workshops (GC Wkshps) 2. J. W. Byers, M. Luby, M. Mitzenmacher and A. Rege “A Digital Fountain Approach to Reliable Distribution of Bulk Data” ACM SIGCOMM Computer Communication Review, vol. 28, no. 4, pp. 56-67, 1998 32