Scalability of Future Mobile Networks Through Network Independent - PowerPoint PPT Presentation

Scalability of Future Mobile Networks Through Network Independent Application Layer Mobility and Context Monitoring Modeling of Adaptive Systems htups://www.mas.wiwi.uni-due.de/en Florian Metzger � 2015/12/10

Motivation QoS and QoE of TCP Streaming in Mobile Networks? 2015/12/10 Modeling of Adaptive Systems Interactions and Optimizations? source: Cisco VNI 2014 2/13 15000 traffic (exabytes per month) traffic type 10000 video web, data, and voip filesharing audio streaming 5000 0 2014 2015 2016 2017 2018 2019 year

Motivation source: Cisco VNI 2014 2015/12/10 Modeling of Adaptive Systems Interactions and Optimizations? QoS and QoE of TCP Streaming in Mobile Networks? 2/13 BSS Core A CS-MGW PSTN A bis I u CS 15000 M c GMSC N c PSTN traffic (exabytes per month) U m BSC MSC BTS traffic type 10000 D C video web, data, and voip UTRAN filesharing Mobile audio streaming HSS Devices G r G c 5000 I ub I ur G n G i SGSN GGSN I u U u S3 G x S6a 0 RNC Node B S4 2014 2015 2016 2017 2018 2019 year S9 E-UTRAN MME PCRF S11 G x G xc S1-MME S1-U S5 S Gi S-GW P-GW X2 eNB U u

Motivation source: Cisco VNI 2014 2015/12/10 Modeling of Adaptive Systems Interactions and Optimizations? QoS and QoE of TCP Streaming in Mobile Networks? 2/13 BSS Core A CS-MGW PSTN A bis I u CS 15000 M c GMSC N c PSTN traffic (exabytes per month) U m BSC MSC BTS traffic type 10000 D C video web, data, and voip UTRAN filesharing Mobile audio streaming HSS Devices G r G c 5000 I ub I ur G n G i SGSN GGSN I u U u S3 G x S6a 0 RNC Node B S4 2014 2015 2016 2017 2018 2019 year S9 E-UTRAN MME PCRF S11 G x G xc S1-MME S1-U S5 S Gi S-GW P-GW X2 eNB U u

Interaction of Complex Systems distinguished only through buffering and 2015/12/10 Modeling of Adaptive Systems on QoS and QoE? Impact of mobile nets and signaling with mobility in mind TCP and TCP-streaming not designed quality level adaptation strategies Fixed protocol stack (TCP+HTTP) Mobile Nets vs TCP Streaming more state (e.g. MME, HSS, PCRF) Other, off-path entities hold and communicate in the net Mobility support, signaling and anchors All user traffic is encapsulated into tunnels the network entities and baseband Control plane is communicated explicitly by 3/13

Interaction Scenario #1 Segment Retrieval and High Latency 2015/12/10 Modeling of Adaptive Systems → Need for testing to find other unexpected interactions! of mobile nets! High latency (and variation) is a common feature to ensure back-to-back transmission Request new segments ahead of time In this case a simple change helps: → Stop-and-wait behavior, overly sensitive to latency when the previous one had arrived New segments were only requested Demonstrates the effects of a bad streaming strategy: HQ video did not complete in time at higher latencies 4/13 120000 90000 video stall duration (s) 60000 low quality standard quality high quality 30000 0 100 1000 additional latency (ms) 150 number of video stalls 100 low quality standard quality high quality 50 0 100 1000 additional latency (ms)

Interaction Scenario #2 High Bandwidths and Buffering 2015/12/10 Modeling of Adaptive Systems Further investigation needed → Bufferbloat TCP won’t back down Packet buffers fill, nothing will be dropped, Negative interaction of TCP and HARQ Excessive buffering in the net Possible explanations: Suspiciously appears above the Stalls at higher bandwidths 5/13 6000 video stall duration (s) 4000 low quality standard quality high quality 2000 radio capacity ( ≈ 80 Mbit s ) 0 10 1000 bandwidth (Mb/s) 90 number of video stalls 60 low quality standard quality high quality 30 0 10 1000 bandwidth (Mb/s)

Interaction Scenario #3 Device Mobility 2015/12/10 Modeling of Adaptive Systems of drops in connectivity Streaming players need to be aware distances → Low buffer and stall events at further 6/13 d/2 d X2 d/2 y SGW/ streaming UE eNodeBs PGW server

Mobile Streaming and Context Factors/Monitoring Mobility as context factor 2015/12/10 Modeling of Adaptive Systems Passing information up the protocol stack, no network assistance so that they can conduct appropriate QoE optimizations themselves Provide an interface with every available context information to applications Requires good predictors, e.g. by deriving information from past patuerns even provide similar or betuer results? Extend the handover blackout to other events of variable lengths, e.g.: Could application-layer mobility in conjunction with context-monitoring Does network-assisted mobility help or hinder in such scenarios for streaming? Subway, metro traffic and tunnels, etc. (Car) traffic tunnels Areas with low radio coverage and insufficient throughput Horizontal and vertical handovers 7/13

“Tunnel” Scenario Optimize segment quality level while still 2015/12/10 Modeling of Adaptive Systems Context-Based Solution advance time is critical Knowledge of upcoming event through context, avoiding stalling 8/13 buffer ahead Compute optimal context-based buffering and QoE quality level selection strategy to ensure best Use context and context predictors in adaptive streaming strategies Prevent stalling, but still do not excessively Buffered video High quality playtime segment Time to play any Time spent to segment (fixed) Low download a segment quality (low layer) No stalling Context- aware HAS (high layer) Standard HAS Proactive lower layer selection Stalling Real time Tunnel warning notification (t adv ) Tunnel start Tunnel end 250 Context − aware HAS Standard HAS Buffer playtime (sec) Optimal HAS 200 150 Tunnel End 100 Tunnel 50 Start 0 0 100 200 300 400 500 600 700 800 Time (sec)

Network Independent Application-Layer Mobility Remove global mobility support 2015/12/10 Modeling of Adaptive Systems Solve remaining issues and provide missing features over-the-top just provide a bare-minimum bit-pipe access Increase scalability/performance by removing most control plane procedures, Provide a trimmed down architecture with only the essentials → Streamlining for future mobile architectures Through Context Monitoring More tricky, but, e.g. adapt existing over-the-top mobility solutions (SIP proxy) VoIP and other real-time communication: network-assisted mobility and just reorder/schedule around Web/HTTP traffic consists of many small objects, could easily completely forgo Is mobility support really necessary for many applications today? “Tunnel” scenario easily transferable from TCP video streaming to other applications 9/13

Streamlining for Future Mobile Architectures High-Level Abstract Architecture 2015/12/10 Modeling of Adaptive Systems 10/13 Access Provider Nets Internet 3GPP base station optional mobilty link 3GPP base station cloud Internet context service Access WiFi AP 3GPP base station 3GPP base station

Streamlining for Future Mobile Architectures Application-layer Mobility Example for Handover 2015/12/10 Modeling of Adaptive Systems 11/13 nodes in range 5G 5G 5G cloud mobile wifi apps other users context service device transceiver #0 transceiver #1 transceiver #2 AP #0 connected context probing request node context selection upload connect early handover mobility notification connected disconnect context upload and request probing node context selection request probing node context selection request

Summary Much potential for (negative) interactions and feedback loops between mobile signaling and (amongst others) streaming E.g. stalling events during handover Requires good understanding and deep investigation Applications could orchestrate their own mobility using context monitoring Mobility scenarios and mobility prediction merit further investigation Goal: Reduce network complexity, increase scalability by moving exiting network features and relinquish control to the application layer Modeling of Adaptive Systems 2015/12/10 12/13

Thanks! E. Liotou, T. Hoßfeld, C. Moldovan, F. Metzger, D. Tsolkas, and N. Passas. “Enriching HTTP Adaptive Streaming with Context Awareness: A Tunnel Case Study”. In: ICC (2016). Under review, submitued in October 2015. F. Metzger, E. Liotou, C. Moldovan, and T. Hoßfeld. “TCP Video Streaming and Mobile Networks: Not A Love Story, But Betuer With Context”. In: Computer Networks Special Issue on Big Data (2016). Under review, submitued in October 2015. F. Metzger. “Evaluating Reliable Streaming in Mobile Networks”. PhD thesis. Apr. 2015. F. Metzger, C. Steindl, and T. Hoßfeld. “A Simulation Framework for Evaluating the QoS and QoE of TCP-based Streaming in an LTE Network”. In: Teletraffic Congress (ITC 27), 2015 27th International . Sept. 2015, pp. 168–176. Qvestions! Modeling of Adaptive Systems 2015/12/10 13/13