Home Network Traffic Measurement

Home Network Traffic Measurement Oscar Olvera-Irigoyen (Orange Labs) Abdesselem Kortebi (Orange Labs) Laurent Toutain (Telecom Bretagne) David Ros (Telecom Bretagne) 1

Outline Introduction Motivations Bandwidth Probing Development Tests Methodology Results and Analysis Conclusions 2

Digital Home Service Provider View 3 Fedosseev, Alex Gabriel Panis, C. MR-239 Broadband Forum Value Proposition for Connected Home Broadband Forum, 2011

Digital Home End User View Reliability Simplicity Comfort Nomadic service capabilit y 4

U.S. Market Size for Connected Home Services 5 Fedosseev, Alex Gabriel Panis, C. MR-239 Broadband Forum Value Proposition for Connected Home Broadband Forum, 2011

Article (Pierce1966) Pierce, J. R. New means of communication Spectrum, IEEE, 1966, 3, 62-65 6

Article (Hatori1986) Hatori, M.; Mokuno, K.; Lida, K.; Ochiai, R. & Horie, T. Home Informatization and Standardization of Home Bus Consumer Electronics, IEEE Transactions on, 1986, CE-32, 542-549 7

Article (Armbruster1986) Armbruster, H. Applications of Future Braod-Band Services in the Office and Home Selected Areas in Communications, IEEE Journal on, 1986, 4, 429-437 8

Beyond Moore's Law: Internet Growth Trends Roberts, L. G. Beyond Moore's Law: Internet Growth Trends Computer, IEEE Computer Society Press, 2000, 33, 117-119 9

Beyond Moore's Law: Internet Growth Trends (cont.) 10

Broadband penetration (in %) is still rising across Europe, the US and Japan Edwards, C. Bringing broadband home Engineering Technology, 2011, 6, 74-75 11

Ecosystem Heterogeneous ecosystem Home Networks have experimented an spectacular complexification: a) New convergent services b) Terminal Diversity c) Hybrid network devices ISP PLC Plug Gateway Power Line Communication flows with conflictive network resource requirements networking constraints Hybrid Extender Wired Ethernet QoS & Network Management Architectures (Difserv, TR-69, UPnP, DLNA, IEEE P1905.1, etc.) QoS metrics Wireless Potentialize Quality of experience 12

networking & QoS in home networks (some names) Differentiated Services in the Internet Home Networking Architecture for IPv6 G.9960 G.9961 TR-58, TR-69, TR-143 P1905.1 QoS QoS 13

QoS Architecture s for HNs 14

General Network/Application Control Loop and Performance Monitoring Passive measuremen t and Probing 15

Cross Layer Available Bandwidth Probing cross traffic Server Probing sequences Bandwidth in Home Networks delay induced to probing sequences Client Based on detecting delay distribution of probing Agnostic to lower layers Classic open source tools (IGI/PTR, Pathchirp y Pathload) : slow and inaccurate in HNs. 16

What Metrics do we use to monitor those HN Paths? Link costs Delay Delay variation Packet losses Capacity Throughput Available bandwidth 17

Example: cross layer available bandwidth probing Available bandwidth: potential transfer resource for convergent services Real time Interactive Background... Available Bandwidth A C Capacity Home Network Path Characterize Network Paths Quality Crucial for Preventive & Corrective Diagnosis 18

Example: cross layer available bandwidth probing (cont.) OK available bandwidth PLC 3D IPtv ISP Home Gateway Hybrid Not enough available bandwidth Adimission Control Wireless Enough available bandwidth? 3D IPtv Preventive Diagnosis 19

Example: cross layer available bandwidth probing (cont.) OK available bandwidth hybrid ISP Home Gateway Not enough available bandwidth PLC OK available bandwidth Best Path Selection Enough available bandwidth? PLC 3D IPtv VoIP Preventive Diagnosis 20

Example: cross layer available bandwidth probing (cont.) Fluctuations of SNR, Capacity & Available Bandwidth ISP Home Gateway Cross traffic IPtv Localize Failures Corrective Diagnosis 21

Measuring with a reference Reference 22

The problem: measuring with a time-variant reference Reference 23

Some home network characteristics Links that introduce variable PHY and MAC overhead Heterogeneous node and terminal performance Home Network Paths that are the last element of the delivery service chain and that can strongly impact the end user Need of accurate and low overhead monitoring (Soft State) 24

lightweight TCP Iperf probing period < 1s Iperf default probing period = 10s TCP << TCP 25

Test bed for lightweight TCP Iperf SND RCV SND RCV 79Mbs 28Mbs 79Mbs 28Mbs Iperf (TX) HPAV (d) AP Iperf (RX) Iperf (TX) HPAV Wireless Bridge (e) Iperf (RX) Ethernet 100Mbs Ethernet 1000Mbs AC Power Line Wireless link Bandwidth Probes Cross Traffic 26

Example Inter-MAC best path selection + lightweight TCP Iperf IPTV Server PLC 802.11a IPTV 1 IMN 1 IMN 2 (a) IPTV 2 Spirent Instrument RX 1 TX RX 2 PLC 802.11a IMN 1 IMN 2 Ethernet 100Mbs (b) AC Power Line Wireless link 27

Inter-MAC best path selection with and without lightweight TCP Iperf 28

Conclusion & Future Work HN monitoring based on simple but representative metrics HN monitoring mapping QoS to QoE or KPIs Simplify architectural designs, network operations and maintenance Using probing is promissory in heterogeneous HNs Probing alleviate passive measurement blind spots Probing require deep study in different time and space scales and diversity of traffic and QoS control mechanisms 29

Home Network Home Network Home Network Internet will essentially be an invisible global infrastructure serving as a global nervous system for the peoples and processes of this planet. Kleinrock, L. 2003 30 Home Network

THANK YOU! QUESTIONS? 31

Probing using gaps (e. g. Wbest) When A tends to decrease g o tends to increase with respect to g i, X X P P P g i Probing Packet Gap Available Bw GAP Model g A = min C 1 o g g i P X P X P Saturation Regime i g o P X Probe Packets Cross Traffic Packets LANMAN 2011, Chapel Hill, October 2011 32

Probing using rates (e.g. PTR, Pathchirp) When A tends to decrease R o, also, tends to decrease with respect to R i. Then R i will be near of A on the onset point of congestion. P X R P X i Probing Rate P If If Available Bw RATE Model R i A min Ri = R R A R > R i > min i P P P X X o o Ro Congestion onset P X Probe Packets Cross Traffic Packets LANMAN 2011, Chapel Hill, October 2011 33

Probing using TCP (e.g. Iperf, Netperf) When the TCP segments tends to the Max Throughput the congestion onset can be detected by 3Dupacks congestion errors and these segments tend also to ocupate the Available bandwidth X S X Segmento TCP Max TCP Troughput Congestion Signal 3Dupacks Max. Throughput for short RTTs and low error rate MaxTCPthroughput Available bandwidth S X X Congestion onset LANMAN 2011, Chapel Hill, October 2011 34

Analysis Tool Performance Criteria Pathchirp Wbest IGI/PTR Iperf Accurate over Ethernet Single bottleneck YES YES YES YES Support multiple bottlenecks YES NO YES YES Adaptative probing to the load NO NO YES YES Good resolution with Low load NO NO NO YES Good resolution with high load NO NO YES YES Resilience to multiple access mechanisms on hybrid paths YES NO NO YES Easy configuration NO YES YES YES Fast sampling on hybrid paths NO YES NO YES Low probing intervals variability YES YES NO YES Low probing overhead YES YES NO NO Global error (95% confidence interval) 33% 40% 29% 15% Average probing interval (s) 2.0 0.4 2.6 0.8 Probing overhead in the narrow link 8% 7% 20% 56% Polling overhead using 5s of Tpo 3% 1% 6% 9% 35

Our contribution Use Iperf in lightweight TCP mode to estimate available bandwidth T pr (Iperf lightweight TCP mode) < 1s << T pr (Iperf default brute force) = 10s Periodic Measurement Intervals T : Test Interval T T T po : Polling Interval (reporting) po T pr T pr : Probing Interval (efective sending of probes) Performance comparison of Iperf vs. Pathchirp, IGI/PTR and Wbest accuracy, probing intervals and overhead using our java test interface 36

Terminal sales evolution to the horizon 2012 MICHAUD, L. Digital Home Entertainment Tendances, Marchés & Prévisions, 2011-2015 2011 37

Motivations The residential market of communications has evolved with a constant demand of service guaranties 38