: So)ware- Defined Networking Mobile Offloading Architecture Junguk Cho, Binh Nguyen, Arijit Banerjee, Robert Ricci, Jacobus Van der Merwe, and Kirk Webb 1/59
MoQvaQon Mobile Network Game server enbs PGW Internet Radio delay Core network delay WAN delay (PGW<- > web server) 2/59
MoQvaQon LTE/EPC Mobile Network Game server enbs PGW Internet Radio delay Short delay Core network delay WAN delay (PGW<- > web server) 3/59
MoQvaQon LTE/EPC Mobile Network Game server enbs PGW Internet Radio delay Core network delay Dominant WAN delay (PGW<- > web server) 4/59
MoQvaQon LTE/EPC Mobile Network Game server enbs PGW Internet Radio delay Core network delay WAN delay (PGW<- > web server) Distance 5/59
MoQvaQon LTE/EPC Mobile Network Game server enbs PGW Internet Radio delay Core network delay WAN delay (PGW<- > web server) Hierarchical rou?ng 6/59
Hierarchical RouQng PGW UE enb Game server Internet The number of enodebs >>>>>>>>>>> The number of S/PGW 7/59
MoQvaQon LTE/EPC mobile network is s?ll not enough for delay- sensi?ve applica?ons like online gaming 8/59
PotenQal SoluQon : Offloading Game server PGW UE enb Move Internet 9/59
PotenQal SoluQon : Offloading Game server PGW UE enb Internet Removing delay between PGW and Internet 10/59
PotenQal SoluQon : Offloading Game server PGW UE enb Internet Allevia?ng delay from hierarchical rou?ng 11/59
Goals Provide mobile offloading architecture with traffic offloading and sonware- define network 12/59
Goals Provide mobile offloading architecture with traffic offloading and sonware- define network SoNware- Define Network(SDN) - Possible fine- grained traffic control on demand - SelecQvely offload traffic based on flow rules 13/59
Goals However, approaches are not new 14/59
Key contribuqons Transparent to exis?ng LTE/EPC mobile network No requirement to modify LTE/EPC architecture and 3gpp standard since it is not easy to change them Built working prototype It supports traffic offloading even when handover happens 15/59
Sweet Spot for locaqon of offloading servers Related to business between mobile carrier and service provider Cost, delay and coverage effecqve locaqon for offloading servers 16/59
Near enodebs LTE/EPC Mobile Network Game server enbs PGW Internet Radio delay Core network delay WAN delay (PGW<- > web server) ~ thousands of enbs Best latency A lot of deployment cost 17/59
Near PGWs LTE/EPC Mobile Network Game server enbs PGW Internet Radio delay Core network delay WAN delay (PGW<- > web server) ~ low tens of PGW Small deployment cost Worst latency 18/59
Regional aggregaqon points LTE/EPC Mobile Network Regional aggrega?on point Game server enbs PGW Internet Radio delay Core network delay WAN delay (PGW<- > web server) ~ 100 150 regional aggregaqon points - It covers a large metropolitan area - Reasonable loca?on in terms of cost and delay 19/59
Regional aggregaqon points LTE/EPC Mobile Network Regional aggrega?on point Cloud Game server enbs PGW Internet Radio delay Core network delay WAN delay (PGW<- > web server) 20/59
Offloading Infrastructure Regional AggregaQon Point SDN PGW Internet enb Offload Cloud MME 21/59
Architecture Controller Regional AggregaQon Point SDN Monitor DB PGW Internet enb Offload Cloud MME 22/59
On- demand use case Use Cases SubscripQon use case 23/59
LTE/EPC Mobile Network On- demand Use Case Regional AggregaQon Point enb MME 24/59
On- demand Use Case Controller Regional AggregaQon Point SDN Monitor DB enb MME Offload Cloud 25/59
Extract Info from Akach Event 26/59
LTE/EPC Mobile Network On- demand Use Case Regional AggregaQon Point enb PGW 27/59
LTE/EPC Mobile Network On- demand Use Case Regional AggregaQon Point Game Frontend Server Internet enb PGW Offload Cloud 28/59
LTE/EPC Mobile Network On- demand Use Case Regional AggregaQon Point Game Frontend Server Internet enb PGW Game Backend Server Offload Cloud 29/59
On- demand Use Case Controller Regional AggregaQon Point Monitor DB Game Frontend Server SDN Internet enb PGW Offload Cloud 30/59
On- demand Use Case Controller Regional AggregaQon Point Monitor DB Game Frontend Server SDN Internet enb PGW Offload Cloud 31/59
On- demand Use Case Controller Regional AggregaQon Point Monitor DB Game Frontend Server SDN Internet enb PGW Offload Cloud 32/59
On- demand Use Case Controller Regional AggregaQon Point Monitor DB Game Frontend Server SDN Internet enb PGW Offload Cloud 33/59
On- demand Use Case Controller Regional AggregaQon Point Monitor DB Game Frontend Server SDN Internet enb PGW Offload Cloud 34/59
On- demand Use Case Controller Regional AggregaQon Point Monitor DB Game Frontend Server SDN Internet enb PGW Offload Cloud Game Backend Server 35/59
On- demand Use Case Controller Regional AggregaQon Point Monitor DB Other services SDN Internet enb PGW Offload Cloud Game Backend Server 36/59
Components SDN Controller Monitor 37/59
SDN SDN Forward traffic based on flow- entries Traffic offloading evaluaqon GTP Tunnel en/decapsulaqon 38/59
GPRS Tunnel Protocol (GTP) enb GTP Tunnel SDN PAYLOAD INNER IP ( src :UE IP, dst : Server IP ) GTP (TEID : TEID ) UDP (dst port : 2125 ) OUTER IP ( src : enb IP, dst : IP ) L2 L1 39/59
Two Possible Paths for Uplink Traffic enb GTP Tunnel SDN PAYLOAD INNER IP ( src :UE IP, dst : Server IP ) GTP (TEID : TEID ) UDP (dst port : 2125 ) OUTER IP ( src : enb IP, dst : IP ) L2 L1 Game Backend Server 40/59
Match Rule for Uplink Traffic enb GTP Tunnel SDN PAYLOAD INNER IP ( src :UE IP, dst : Server IP ) GTP (TEID : TEID ) UDP (dst port : 2125 ) OUTER IP ( src : enb IP, dst : IP ) L2 L1 Match fields 41/59
Uplink Traffic RedirecQon enb GTP Tunnel SDN Flow Table Match GTP (TEID : TEID ) & INNER IP ( src :UE IP, dst : Server IP ). AcQon Ac?on 1.Decapsulate GTP, UDP, OUTER IP 2.Forward it to game backend server. Game Backend Server PAYLOAD INNER IP ( src :UE IP, dst : Server IP ) L2 L1 42/59
Uplink Traffic enb GTP Tunnel SDN GTP Tunnel PAYLOAD INNER IP ( src :UE IP, dst : Server IP ) GTP (TEID : TEID ) UDP (dst port : 2125 ) OUTER IP ( src : enb IP, dst : IP ) L2 L1 PAYLOAD INNER IP ( src :UE IP, dst : Server IP ) GTP (TEID : TEID ) UDP (dst port : 2125 ) OUTER IP ( src : enb IP, dst : IP ) L2 L1 43/59
Downlink Traffic SDN GTP Tunnel Game Backend Server 44/59
Downlink Traffic SDN Game Backend Server PAYLOAD INNER IP ( src :UE IP, dst : Server IP ) L2 L1 Match fields 45/59
Downlink Traffic enb SDN PAYLOAD INNER IP ( src : Server IP, dst : UE IP ) GTP (TEID : enodeb TEID ) UDP (dst port : 2125 ) OUTER IP ( src : IP, dst : enb IP ) L2 L1 Flow Table Match INNER IP ( src :UE IP, dst : Server IP ). AcQon Ac?on 1. Encapsulate GTP, UDP, OUTER IP 2. Forward it to enb. Game Backend Server 46/59
Downlink Traffic enb SDN GTP Tunnel PAYLOAD INNER IP ( src : Server IP, dst : UE IP ) GTP (TEID : enodeb TEID ) UDP (dst port : 2125 ) OUTER IP ( src : IP, dst : enb IP ) L2 L1 PAYLOAD INNER IP ( src : Server IP, dst : UE IP ) GTP (TEID : enodeb TEID ) UDP (dst port : 2125 ) OUTER IP ( src : IP, dst : enb IP ) L2 L1 47/59
ImplementaQon LTE/EPC testbed: OpenEPC LTE/EPC so)ware SDN : open Vswitch 2.0 Extending GTP evaluaqon, en/decapsulaqon funcqons by using vport abstracqon mechanism Controller : Ryu controller Extending Ryu API for GTP flow management Monitor : Tshark DetecQng akach and detach events ExtracQng informaqon from events and storing them to DB. 48/59
PhantomNet Testbed tunneled traffic traffic without tunneling Internet Controller & Monitor UE enodeb SDN MME PGW Offloaded Server 49/59
EvaluaQon SDN overhead End- to- End RTT improvement We use ping for evaluaqon 50/59
SDN Overhead Measure overheads from addiqonal funcqonaliqes in SDN GTP evaluaqon GTP en/decapsulaqon Compare SDN with naqve open Vswitch SDN enb Packet In In_port Matches & AcQons Out_port Packet Out or Oflloading server Elapsed Time 51/59
Uplink SDN Overhead enb SDN Packet In In_port Matches & AcQons Overheads from 1. GTP evalua?on 2. GTP decapsula?on Out_ port Packet Out Game Backend Server 52/59
Uplink SDN Overhead enb Packet In In_port SDN Matches & AcQons Out_port Packet Out Overhead from GTP evalua?on 53/59
Uplink SDN Overhead Processing time in each port(us) 12 10 8 6 4 2 0 ETH input port ETH output port GTP Decap GTP eval (CLOUD) GTP eval GTP eval port GTP decap port NATIVE OVS Compared with Na?ve OVS - ( to CLOUD) : 5.3 us - ( to ) : 1.9 us 54/59
Downlink SDN Overhead enb Packet Out In_port SDN Matches & AcQons Out_ port Overhead from GTP decapsula?on Packet In Game Backend Server 55/59
Downlink SDN Overhead Processing time in each port(us) 8 7 6 5 4 3 2 1 0 ETH input port ETH output port GTP Encap (CLOUD) GTP encap port NATIVE OVS Compared with Na?ve OVS - ( from CLOUD) : 2.1 us 56/59
Feasibility and LimitaQon of The overall overhead due to processing GTP func?onali?es in SDN is not high compared to end- to- end delay in LTE (~70ms) May have scalability issue in SDN Explore scalability in SDN Architectural limitaqon Limited support for handover (supporqng Intra- LTE Handover using the X2 interface) 57/59
Conclusion We presented architecture to realize traffic offloading to reduce end- to- end delay No modificaqon of exisqng LTE/EPC mobile network. Show how offloading for selected traffic of subscribed users can realized even when handover happens Prototype realizaqon of architecture in PhantomNet LTE/EPC testbed. 58/59
Thank you! QuesQons? Visit phantomnet.org 59/59