1. Service QoE Monitoring in the Access Network Bart De Vleeschauwer Ghent University – IBBT-IMEC Department of Information Technology Bart.DeVleeschauwer@intec.ugent.be

2. 03.10.2006 — 2 Muse Project > MUlti Service access Everywhere > “The overall objective of MUSE is the research and development of a future, low cost, multi-service access network. The access network should provide secure connectivity between end-user terminals and edge nodes in a multi-provider environment. It should be suited for the ubiquitous delivery of broadband services to every European citizen.” http://www.ist-muse.org > Service QoE is of prime importance > Goal of our contribution to MUSE: a knowledge plane in the access network that is responsible for autonomous QoE management.

3. 03.10.2006 — 3 Outline > Access network overview > Access network services > Knowledge plane Motivation A two layered architecture > Access node monitoring and actions RTP/RTCP monitoring QoE restorative actions > Conclusion

4. 03.10.2006 — 4 Access Network Overview Service Providers Service Edge Aggregation Network Access Node Residential Gateway End Device User Home Network

5. 03.10.2006 — 5 Motivation > Access network services: VOIP Video on demand IPTV High Speed Internet > For all these services, QoE is essential > Packet loss, delay, jitter have huge impact > Goal: Autonomous access network QoE management Monitor the QoE Enable QoE restorative action Proactive and reactive QoE monitoring and restoration

6. 03.10.2006 — 6 A Knowledge plane in the access network Monitoring Plane Knowledge Plane

7. 03.10.2006 — 7 State of the art > Monitoring the access network SNMP (Simple Network Management Protocol) RMON MIBs Raqmon DSLForum – Protocol described in TR-69 – Object models in various working texts IPFIX/Netflow Active monitoring > Monitor data analysis “Sketch-based change detection: methods, evaluation, and applications”, Krishnamurthy, B., Sen, S., Zhang, Y., Chen, Y., 2003 “What’s new: finding significant differences in network data streams”, Cormode, G., Muthukrishnan, S., 2005 Database sliding window techniques > Knowledge plane actions Retransmissions Forward error correction Interleaving

8. 03.10.2006 — 8 Monitoring Plane MP/KP Responsibilities Data Reduction Monitored Data SamplingSliding WindowSketch Knowledge Plane Reduce data to manageable size Anomaly Detection Sketch Test 2ThresholdSketch Test 1 Diagnosis and Solution Alert diagnosis component Detect problem root cause and find solution Analyze data

9. 03.10.2006 — 9 A distributed Knowledge Plane Monitoring Plane Knowledge Plane Monitoring Plane

10. 03.10.2006 — 10 Monitoring Plane Data Reduction Knowledge plane – Monitoring plane interaction Knowledge Plane Active: e.g. generate additional ICMP ping requests Passive: e.g. additional threshold in RMON MIB Anomaly Detection Diagnosis and Solution Additional queries over available data Initialize new anomaly detection modules Additional info might be needed for accurate fault recovery Initiate new monitor probes Anomaly is detected

11. 03.10.2006 — 11 Central role of the access node > Access node: Under control of access network provider Crossing point between aggregation network and last mile + home network All data of the same user passes this point > Ideal place for user/service monitoring and restorative action > Dedicated protocol monitoring Knowledge Plane Monitoring Plane

12. 03.10.2006 — 12 Streaming video services Streaming video service RTP RTCP

13. 03.10.2006 — 13 RTP/RTCP > RFC 3550: RTP: A Transport Protocol for Real-Time Applications > Two protocols RTP protocol for data packets RTCP protocol for control traffic > RTP packets contain data Sequence number Timestamps (Sampling instant of first octet in the RTP data packet) > RTCP packets contain control & feedback information

14. 03.10.2006 — 14 RTCP Messages > SDES, source description items > BYE, end of participation > APP, application specific > SR, Sender Report > RR, Receiver Report Report Block

15. 03.10.2006 — 15 Monitor Plane Streaming Video Services Goal: Determine end-to-end QoE Determine Access-node end-device QoE RTP SR RR Track Sequence NumbersTrack SR timesInspect RRs

16. 03.10.2006 — 16 RTP/RTCP loss estimation > Receiver calculates the number of packets it expects and the number it has received in an interval > Fraction lost between two report blocks is reported together with cumulative number of lost packets > Cumulative number lost = # expected - # received > Fraction lost = fraction lost since last RR

17. 03.10.2006 — 17 Access node home network loss detection Access node keeps track of packets that were sent to end-device Access node keeps track of highest sequence numbers in RR Access node keeps track of reported fraction lost / number lost packets Access node can estimate how many packets were lost between access node and end- device

18. 03.10.2006 — 18 RTP/RTCP interarrival Jitter estimation > Interarrival jitter: variance in interarrival time Server End-Device SlSl time > End-to-end interarrival jitter is calculated: Calculate interarrival time between two packets D(k,l) = (R k – R l ) - (S k – S l ) = (R k – S k )- (R l – S l ) Jitter estimation J(k)=J(k-1)+(|D(k,k-1)|- J(k-1) )/16 > Jitter is reported in RR > Do analogous calculations to determine Server – AN jitter > Compare end-to-end jitter (RR) and Server-AN jitter SkSk RlRl RkRk

19. 03.10.2006 — 19 RTP/RTCP RTT estimation > RTCP Report block contains fields for LSR: Last SR timestamp DLSR: Delay Since last SR > This allows us to estimate the RTT whenever a RR is received > Time when SR passes access node: t1 > Time when RR passes access node : t2 > RTT est. = t2-t1-DLSR Sender Access Node End Device DLSR t1 t2 SR RR

20. 03.10.2006 — 20 Access Node QoE Restorative Action > Trigger application specific actions when packet loss is detected Forward error correction Interleaving Intercept retransmission requests at the access node and send retransmissions – Faster retransmission – Less network load > Inform higher layers of QoE decrease > Adapt stream to better suite client specifications > Dynamic content proxying according to an observed demand pattern

21. 03.10.2006 — 21 Conclusion > There is a clear need for access network QoE management > Three goals QoE monitoring QoE problem detection Problem solution > A two layered solution Monitor plane Knowledge plane > A distributed architecture is able to locate/solve QoE decreases at the appropriate location > The access node plays a central role in the distributed knowledge plane

22. 03.10.2006 — 22 Thank you for you attention ! Any Questions?