UCLA ENGINEERING Computer Science A Software Defined Multi-Path Traffic Offloading System for Heterogeneous LTE-WiFi Networks Pengyuan Du * , Qi Zhao * and Mario Gerla Department of Computer Science, UCLA IEEE WoWMoM 2019 6/11/2019 * Co-primary author
UCLA ENGINEERING Computer Science In Memory of Professor Mario Gerla ❑ “The imprint Prof. Gerla made on all our lives is indelible. He was a beloved husband, father, brother to his family, a thought leader and brilliant mind within the global computer science community, a mentor and inspiration to his Masters and PhD students, and a dear friend to us all. May we all honor his spir- it by living life with passion, dedication and open minds. ” 75 th Birthday
UCLA ENGINEERING Computer Science Outline ❑ Introduction ❑ Software-Defined Multi-Path Traffic OPtimization (SD-MPTOP) Traffic Offloading in HetNet ➢ SD-MPTOP Overview ➢ Multi-Commodity Flow Modeling ➢ ❑ SD-MPTOP Implementation ❑ Performance Evaluation ❑ Conclusion
UCLA ENGINEERING Computer Science Introduction ❑ The era of mobile communications Vehicle Mobile phone Home
UCLA ENGINEERING Computer Science Introduction ❑ The explosion of mobile communications [1] ➢ Sevenfold increase in mobile data traffic ➢ Spectrum crisis Mobile data traffic (Exabyte) 60 50 40 30 20 10 0 2016 2017 2018 2019 2020 2021 [1] Cisco. “Cisco Visual Networking Index: Forecast and Methodology, 2016–2021.”, 2018.
UCLA ENGINEERING Computer Science Introduction ❑ Offloading cellular traffic Use of alternative network technologies for data delivery ➢ Infrastructure ➢ Small-cell / WiFi / 3GPP • Opportunistic networks ( aka Mobile ad-hoc networks) ➢ Delay tolerant networks (DTN) • Device-to-Device (D2D) communications • D2D DTN Pico cell Femto cell Macro cell Micro cell Content Multi-home distribution
UCLA ENGINEERING Computer Science Introduction ❑ Heterogeneous infrastructure-based network (HetNet) ➢ Current practice: direct use of WiFi ➢ Our solution: multipath , concurrent use of cellular and WiFi User: bandwidth aggregation • Network operator: centralized management •
UCLA ENGINEERING Computer Science Traffic Offloading in HetNet Internet Internet Internet Internet Internet LTE LTE LTE WiFi LTE WiFi LTE LTE WiFi MPTCP
UCLA ENGINEERING Computer Science Multipath TCP Protocol ❑ Multipath TCP (MPTCP) [1] ➢ Presenting a single TCP connection to the application ➢ Utilize different interfaces underneath ➢ Work over today’s networks [1] Wischik, Damon, et al. "Design, Implementation and Evaluation of Congestion Control for Multipath TCP." NSDI, 2011.
UCLA ENGINEERING Computer Science Traffic Offloading in HetNet Current practice Global view Multipath ❑ Resource allocation problem single path optimal selection user selection Server Internet Internet Internet Internet Capacity HetNet HetNet HetNet HetNet HetNet HetNet HetNet HetNet HetNet HetNet HetNets WiFi WiFi 1 2 3 1 1 1 2 3 3 3 Physical connectivity Devices
UCLA ENGINEERING Computer Science Software Defined Networking ❑ SDN architecture ➢ Decoupling of control and data plane Centralized management • Network function virtualization • Network� Virtualization Well-defined� API Traffic� Other� Routing Application Plane Security Applications Engineering Network� Map� Abstraction Network� Operating� System C ont rol P l l a ane ne Instructions Instructions Instructions Instructions Separation� of� Data� Forwarding and� Control Plane Forwarding Forwarding D a D at t a a P l P l ane a ne Forwarding
UCLA ENGINEERING Computer Science Traffic Offloading in HetNet ❑ Design considerations i. Efficient bandwidth aggregation ii. Service continuity MPT PTCP iii. Computation hardness SD SDN iv. Network management, QoS, Deployment cost ❑ Proposed solution architecture ➢ Multipath TCP (MPTCP) ➢ Software Defined Networking (SDN)
UCLA ENGINEERING Computer Science System Overview 1 1 0 0 0 1 1 1 1 MP Assoc. MCF Solver Config MPTCP Network Controller report SYN/ACK WiFi SDN Switch subflow1 subflow2 subflow1 subflow2 MPTCP Data plane Control plane
UCLA ENGINEERING Computer Science Traffic Offloading in HetNet Conn. Matrix WiFi WiFi LTE LTE MN1 MN1 0 1 0 1 MP Assoc. MN2 MN2 1 1 1 1 . . . . . . . . . . . . . . . . . . MNk MNk 1 0 1 0 Controller Subflow1 Subflow1 MPTCP WiFi Internet . . . MPTCP Subflow1 Subflow1
UCLA ENGINEERING Computer Science Traffic Offloading in HetNet Conn. Matrix WiFi WiFi LTE LTE MN1 MN1 0 1 0 1 MP Assoc. MN2 MN2 1 1 1 1 . . . . . . . . . . . . . . . . . . MNk MNk 1 0 1 0 Controller LTE WiFi MCF Solver MNk RSSI Capacity Net Load MN1 MN1 Bandwidth MPTCP Allocaion WiFi Internet RSSI Capacity Net Load . . . MPTCP
UCLA ENGINEERING Computer Science Traffic Offloading in HetNet LTE MP Assoc. WiFi Controller MCF Solver MNk MN1 MN1 Bandwidth Config Allocaion MPTCP WiFi Internet SDN Switch . . . Switch Configurations MPTCP Action Flows Queue1,Output:3 Src_ip: 10.0.3.0 SDN Switch Queue2,Output:4 Src_ip:10.0.3.1 Src_ip:10.0.2.0 Queue3,Output:3 Queue4,Output:4 Src_ip:10.0.2.1
UCLA ENGINEERING Computer Science Traffic Offloading in HetNet MP Assoc. Controller MCF Solver Config MPTCP WiFi Software-Defined MultiPath Internet Traffic OPtimization (SD-MPTOP) . . . MPTCP
UCLA ENGINEERING Computer Science SD-MPTOP MCF ❑ Multi-Commodity Flow (MCF) Internet min Obj( ) f , n s f n s , BW BW BW 1 2 3 Subject to: Net1 Net2 Net3 Capacity constraint f BW n s , s n f f f 3,3 1,1 1,2 = Requirement constraint f R f n s , n 3,2 s MN1 MN2 MN3 MN4 R R R R 1 2 3 4
UCLA ENGINEERING Computer Science SD- MPTOP MCF (cont’d) ❑ Routing Assignment ➢ Objective function: minimizes the average packet delay f n s , 1 = + where n N min T s f d − n s , n s , BW f f n s , s s n N s n s , s n N s link l oa d Load balancing Propagation delay or network cost Service rate BW s f Flow rate Link Load n s , n N cost s M/M/1 Queue delay model Network congestion indicator [1] BW f s f n s , n s , n N s [1] Fratta, Luigi, Mario Gerla, and Leonard Kleinrock. "The flow deviation method: An approach to store ‐ and ‐ forward communication network design."
UCLA ENGINEERING Computer Science SD- MPTOP MCF (cont’d) ❑ Example Objective function: L f n s , 1 = + n N min T s f d − n s , n s , BW f f , n s s s n N s n s , s n N s link load Internet 1. Multipath 1. Single path 1. Multipath 2. Load balanced 2. LTE underutilized 2. WiFi congested BW =6Mbps BW =3Mbps 2 1 3. WiFi congested LTE WiFi LTE LTE LTE MN1 MN1 MN1 MN2 MN2 MN2 MN3 MN3 MN3 6M 6M 6M MN1 MN2 MN3 WiFi WiFi WiFi <1.00 <1.00 3.50 3M 3M R =0.5Mbps R =4Mbps R =1Mbps 3M 1 2 3 y 4-y <4.00 1.00 0.50 <0.50 x 0.5-x 0.44 0.06 Bandwidth Bandwidth Bandwidth Allocation of Allocation of Allocation of SPTCP MPTCP SD-MTOP
UCLA ENGINEERING Computer Science SD-MPTOP MCF Solution ❑ The Flow Deviation Method (FDM) ❑ Minimum allocation interruption f n s , 1 2 = + + − n N min T s f d f f − n s , n s , n s , n s , BW f f n s , s s n N s n N s n s , s s n N s link int errruption lo d a ❑ Quick start Use previous allocation as the initial feasible solution ➢ ❑ Max-min fairness MCF could be infeasible ➢ Utilizing all the bandwidth in a fair manner ➢ Binary search the maximum demand that nearly saturates the capacity ➢
UCLA ENGINEERING Computer Science SD-MPTOP Implementation MP Assoc. Config MCF Solver Controller draco.cs.ucla.edu Controller WiFi Linux TC wmediumd eth0 wlan0 eth0 wlan0 Mobility support MPTCP Linux kernel Controller MPTCP
UCLA ENGINEERING Computer Science SD-MPTOP Evaluation Handover scenario Static scenario Mobile scenario
UCLA ENGINEERING Computer Science Static scenario Parameters N1 5 N2 5 BW wifi 4Mbps 6Mbps BW LTE Delay wifi 10ms Delay LTE 50ms
UCLA ENGINEERING Computer Science Handover scenario Parameters Static, mobile 1,1 BW wifi 6Mbps BW LTE 8Mbps 10ms Delay wifi Delay LTE 50ms Req 3Mbps
UCLA ENGINEERING Computer Science Mobile scenario Mobile Static Parameters Static 5 5 mobile BW wifi 6Mbps BW LTE 8Mbps Delay wifi 10ms Delay LTE 50ms Req 2Mbps
Recommend
More recommend
