towards detecting differential qoe
play

Towards Detecting Differential QoE David Choffnes, Northeastern - PowerPoint PPT Presentation

Towards Detecting Differential QoE David Choffnes, Northeastern (joint work with SBU) Supported by a Google Grant Traffic differentiation 2 Traffic differentiation selectively changing the performance of network traffic Reasons for


  1. Towards Detecting Differential QoE David Choffnes, Northeastern 
 (joint work with SBU) Supported by a Google Grant

  2. Traffic differentiation 2 Traffic differentiation selectively changing the performance of network traffic � Reasons for differentiation: ¤ traffic engineering ¤ bandwidth management ¤ business reasons

  3. Traffic differentiation 2 Traffic differentiation selectively changing the performance of network traffic � Reasons for differentiation: ¤ traffic engineering ¤ bandwidth management ¤ business reasons Do certain types of network traffic receive better (or worse) QoE ?

  4. Motivation: Differentiation in the wild 3 � Tested in early 2015 ISP YouTube Netflix Spotify • m: content modified on the fly Verizon m m m - - - Tmobile • p: translucent proxies change m m m ATT connection behavior m m m Sprint Boost m m m BlackWireless 60% - - H2O 37% 45% 65% SimpleMobile 36% - - NET10 p p p

  5. Motivation: Differentiation in the wild 3 � Tested in early 2015 ISP YouTube Netflix Spotify • m: content modified on the fly Verizon m m m - - - Tmobile • p: translucent proxies change m m m ATT connection behavior m m m Sprint Boost m m m BlackWireless 60% - - H2O 37% 45% 65% SimpleMobile 36% - - NET10 p p p

  6. Motivation: Differentiation in the wild 3 � Tested in early 2015 ISP YouTube Netflix Spotify • m: content modified on the fly Verizon m m m - - - Tmobile • p: translucent proxies change m m m ATT connection behavior m m m Sprint Boost m m m BlackWireless 60% - - H2O 37% 45% 65% SimpleMobile 36% - - NET10 p p p

  7. Motivation: Differentiation in the wild 3 � Tested in early 2015 � Again in late 2015: No observed differences ISP YouTube Netflix Spotify • m: content modified on the fly Verizon m m m - - - Tmobile • p: translucent proxies change m m m ATT connection behavior m m m Sprint Boost m m m BlackWireless 60% - - H2O 37% 45% 65% SimpleMobile 36% - - NET10 p p p

  8. Key Questions 4 What do you test for differentiation? ¤ How do you generate traffic? ¤ What triggers differentiation? How can you tell if there is differentiation? ¤ How do you do a controlled experiment? ¤ How do you rule out other reasons for differential service? What is the impact on QoE? ¤ How do you map observed degradation to QoE? ¤ How do you scale this to arbitrary applications?

  9. How do you test for differentiation? 5 � What triggers differentiation? ¤ We don’t know � They might trigger on ¤ IP addresses ¤ ports ¤ payload signatures ¤ total number of connections ¤ total bandwidth ¤ time of day (This is consistent with online manuals for DPI boxes)

  10. How do you test for differentiation? 6 � What triggers differentiation? ¤ We tested using carrier-grade DPI boxes � They might trigger on ¤ IP addresses ¤ ports ¤ payload signatures ¤ total number of connections ¤ total bandwidth ¤ time of day

  11. What triggers differentiation? 7 � HTTP ¤ Host: and GET fields, typically regex ¤ Examples: youtube, facebook, netflix � HTTPS ¤ Server cert is sent in plaintext ¤ Searches on SNI , CN fields � Other protocols ¤ Can identify Skype using some knowledge of handshake format ¤ Open question : how to reverse engineer classifiers? � What is not important ¤ IP addresses don’t seem to matter!

  12. Triggering differentiation 8

  13. Triggering differentiation 8

  14. Triggering differentiation 8 end user

  15. Triggering differentiation 8 end user

  16. Triggering differentiation 8 end user

  17. Triggering differentiation 8 VPN TUNNEL VPN proxy end user

  18. Triggering differentiation 8 Record: VPN TUNNEL VPN proxy end user

  19. Triggering differentiation 8 Record: VPN TUNNEL VPN proxy end user Replay:

  20. Triggering differentiation 8 Record: VPN TUNNEL VPN proxy end user Replay: end user

  21. Triggering differentiation 8 Record: VPN TUNNEL VPN proxy end user Replay: Replay server end user

  22. Triggering differentiation 8 Record: VPN TUNNEL VPN proxy end user Replay: VPN TUNNEL Replay server end user

  23. Triggering differentiation 8 Record: VPN TUNNEL VPN proxy end user Replay: VPN TUNNEL Replay server end user

  24. Triggering differentiation 8 Shaper Record: VPN TUNNEL VPN proxy end user Replay: VPN TUNNEL Replay server end user

  25. Triggering differentiation 8 Shaper Record: VPN TUNNEL VPN proxy end user Replay: VPN TUNNEL Replay server end user

  26. Triggering differentiation 8 Shaper Record: VPN TUNNEL VPN TUNNEL VPN proxy end user Replay: VPN TUNNEL Replay server end user

  27. How to tell if there is differentiation? 9 � Developed and validated new detection technique ¤ Back-to-back tests of same trace ¤ Includes VPN and random payload (but not ports) ¤ Send only at recorded rate (this is not a speed test) ¤ Statistical tests to reliably find differentiation

  28. How to tell if there is differentiation? 9 � Developed and validated new detection technique ¤ Back-to-back tests of same trace ¤ Includes VPN and random payload (but not ports) ¤ Send only at recorded rate (this is not a speed test) ¤ Statistical tests to reliably find differentiation KS Test CDF statistic Throughput (KB/s)

  29. How to tell if there is differentiation? 9 � Developed and validated new detection technique ¤ Back-to-back tests of same trace ¤ Includes VPN and random payload (but not ports) ¤ Send only at recorded rate (this is not a speed test) ¤ Statistical tests to reliably find differentiation w a KS Test CDF CDF statistic Area Test = a/w Throughput (KB/s) Throughput (KB/s)

  30. Mapping to QoE 10 � Currently measure throughput, loss, delay, jitter ¤ Some clear mappings to video streaming bitrate ¤ Fairly clear mapping to VoIP ¤ … but unclear how it impacts user-perceived performance � Key challenge: Applications are adaptive , servers dynamic ¤ Users may not perceive impairment � Current focus: Expose QoE metrics ¤ YouTube does this, Netflix does not ¤ What is the right format? ¤ How do we know if users notice?

  31. Summary 11 � Differentiation and its impact on QoE important to solve � Detecting differentiation requires care ¤ What you send (trigger it), how you send it (avoid it) ¤ Detection approach should be resilient to noise � Need a deeper understanding of impact on QoE ¤ Adaptive applications pose a challenge ¤ Potential approach is combination of ¤ Exposing more QoE data from applications ¤ Building better models to map QoS to QoE

  32. Questions? 
 http://dd.meddle.mobi Check out our app!

Recommend


More recommend