Measuring YouTube Content Delivery over IPv6 Vaibhav Bajpai TU Munich
Motivation Methodology Success Rate
IETF 99 Prague, CZ Published at: SIGCOMM Computer Communication Review July 2017: https://goo.gl/oUJ7Ej
IPv6 Preference TCP connect times Startup Delay Throughput Stall Events
Joint work with
Recommendations Q/A
Saba Ahsan Aalto University, Helsinki, Finland Jörg Ott TU Munich, Germany Jürgen Schönwälder Jacobs University Bremen, Germany July 2017 1 / 18
Motivation Google IPv6 Adoption
▶ IPv6 contributes ∼25% [1] of traffic within Comcast. ▶ Swisscom reports ∼60% [1] of IPv6 traffic is YouTube. ▶ IPv6 traffic largely dominated by YouTube [2].
20% 15% 10% 5% 0%
Motivation Methodology Success Rate 2009 2010 2011 2012 2013 2014 2015 2016 2017
shaded region represents the duration of the longitudinal study.
Do users experience benefit (or suffer) from YouTube streaming over IPv6?
IPv6 Preference TCP connect times Startup Delay Throughput
NETWORK TYPE RESIDENTIAL NREN / RESEARCH BUSINESS / DATACENTER OPERATOR LAB IXP
RIR
∼100 dual-stacked SamKnows probes (∼66 different origin ASes)
RIPE ARIN APNIC AFRINIC LACNIC
# 78 10 08 04 01
Stall Events Recommendations Q/A
# 60 29 10 01 01
2 / 18
Research Contribution Motivation Methodology
▶
HE (RFC 6555) makes clients to prefer streaming YouTube videos over IPv6.
▶
Observed performance (both in terms of latency and throughput) over IPv6 is worse.
▶
Stall rates are low, bitrates that can be reliably streamed are comparable.
▶
When a stall occurs, stall durations over IPv6 are higher.
▶
Worse performance is due to GGC nodes that are IPv4-only.
Success Rate IPv6 Preference TCP connect times Startup Delay Throughput Stall Events Recommendations Q/A
This is the first study to measure YouTube content delivery over IPv6
3 / 18
Methodology | Selection of YouTube Videos Motivation Methodology
▶
Using YouTube v3 API [3].
▶
Video Selection Criteria:
Success Rate IPv6 Preference TCP connect times
1. Video duration > 60s. 2. Available in Full HD. 3. No regional restrictions. ▶
List updated every 12h.
▶
Probes daily pull the list.
▶
The test supports non-adaptive and step-down playout modes only.
▶
Results are biased our vantage points (centered largely around EU, US and JP).
Startup Delay Throughput Stall Events Recommendations Q/A
4 / 18
Success Rate Motivation
Success Rate
The test executes once every hour (over both AF).
Methodology Success Rate
IPv6 (102)
IPv6 Preference
1.0 0.8 0.6 0.4 0.2 0.0
['14 - '17]
▶
Number of successful iterations to total iterations. CCDF
▶
1.0 0.8 0.6 0.4 0.2 0.0
IPv4 (102)
TCP connect times Startup Delay Throughput Stall Events Recommendations Q/A
80%
85%
90%
95%
▶
99% of probes achieve success rate of more than 94% over IPv4.
▶
97% of probes achieve success rate of more than 94% over IPv6.
▶
Slightly lower success rates over IPv6 due to network issues closer to probes.
100%
5 / 18
IPv6 Preference Motivation
1.0 0.8 0.6 0.4 0.2 0.0
Methodology
['14 - '17]
CCDF
IPv6 Preference Web (871) Audio (871) Video (871) 95%
96%
97%
98%
99%
100%
Success Rate IPv6 Preference TCP connect times Startup Delay Throughput Stall Events Recommendations
▶
RFC 6724 [4] makes apps prefer connections made over IPv6.
▶
RFC 6555 [5] allows apps to fallback to IPv4 when IPv6 connectivity is bad.
▶
TCP connections over IPv6 are preferred at least 97% of the time.
Q/A
Clients prefer streaming YouTube videos over IPv6 6 / 18
TCP connect times Motivation Methodology
1.0 0.8 0.6 0.4 0.2 0.0
Success Rate
['14 - '17]
CDF
TCP connect times Web (6.5M) Audio (6.5M) Video (6.5M)
−100
−75
−50
−25 ∆t (ms)
0
25
50
IPv6 Preference TCP connect times Startup Delay Throughput Stall Events Recommendations Q/A
▶
63% of a/v streams (and 72% of the web connections) are slower over IPv6.
▶
14% of a/v streams are at least 10 ms slower over IPv6.
7 / 18
TCP connect times Motivation
∆t (ms)
∆t (ms)
TCP Connect Times
Methodology
0
Success Rate IPv6 Preference
Web −5
0.0 −0.1 −0.2 −0.3 −0.4
TCP connect times
TCP Connect Times
Startup Delay Throughput
Audio Video
Stall Events Recommendations
Jan 2015
Jul
Jan 2016
Jul
Jan 2017
Jul
▶
TCP connect times consistently higher over IPv6 and have not improved over time.
▶
TCP connect times towards the webpage worse over IPv6 than towards media servers.
Q/A
8 / 18
Sequence Diagram (contd.) Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Throughput Stall Events Recommendations Q/A
9 / 18
Startup Delay Motivation Methodology
1.0 0.8 0.6 0.4 0.2 0.0
Success Rate
['14 - '17]
CDF
Startup Delay 6.5M
−5
−4
−3
−2 −1 ∆s (s)
0
1
IPv6 Preference TCP connect times Startup Delay Throughput Stall Events Recommendations Q/A
▶
80% of the samples are slower over IPv6.
▶
Half of the samples are at least 100 ms slower over IPv6.
10 / 18
Startup Delay Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Throughput Stall Events Recommendations Q/A
▶
Prebuffering durations are ∼25 ms higher over IPv6.
▶
Startup delays are ∼100 ms higher over IPv6.
▶
Initial interaction with the web server makes startup delay worse over IPv6. 11 / 18
Sequence Diagram (contd.) Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Throughput Stall Events Recommendations Q/A
12 / 18
Throughput Motivation
Video (6.5M) Audio (6.5M)
−4
∆tp (MB/s)
['14 - '17]
CDF
Throughput 1.0 0.8 0.6 0.4 0.2 0.0
−2
0 ∆tp (MB/s)
2
4
Success Rate IPv6 Preference TCP connect times Startup Delay Throughput
Throughput
0.2 −0.1 −0.4 −0.7 −1.0
Methodology
Stall Events Recommendations
Video Jan 2015
Jul
Jan 2016
Jul
Audio Jan 2017
Q/A Jul
▶
80% of video and 60% audio samples achieve lower throughput over IPv6.
▶
The throughput is consistently lower over IPv6, but it has improved over time. 13 / 18
Sequence Diagram (contd.) Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Throughput Stall Events Recommendations Q/A
14 / 18
Stall Rates Motivation
1.0 0.8 0.6 0.4 0.2 0.0
Methodology Success Rate IPv6 Preference
['14 - '17]
CDF
Stall Rate
IPv6 (102)
1.0 0.8 0.6 0.4 0.2 0.0
IPv4 (102)
TCP connect times Startup Delay Throughput Stall Events Recommendations Q/A
0%
5%
10%
15%
20%
25%
30%
▶
90% of the probes witness less than 1% stall rate over both address families.
▶
Bitrates reliably streamed is also comparable over both address families. 15 / 18
Stall Durations Motivation Methodology
1.0 0.8 0.6 0.4 0.2 0.0
Success Rate
1.7K −60
−40
−20 0 ∆st (s)
20
40
['14 - '17]
CDF
Stall Durations
60
IPv6 Preference TCP connect times Startup Delay Throughput Stall Events Recommendations Q/A
▶
80% of samples experience stall durations that are at least 1s longer.
16 / 18
Recommendations Motivation
▶
Update RFC 6555 with a lower HE timer value. ▶
We have shown [6] that reducing HE timer value to 150 ms (from 300 ms) helps.
Methodology Success Rate IPv6 Preference
0.10%
W6LD
0.15%
W6D
0.20%
6to4/Teredo Decline
TCP connect times Startup Delay Throughput
0.05%
Stall Events
0.00% 2009 2010 2011 2012 2013 2014 2015 2016 2017
Recommendations Q/A
▶
ISPs should put latency as a first-class citizen.
▶
ISPs should ensure GGC nodes are dual-stacked. ▶
Request an IPv6 prefix allocation from Google.
17 / 18
Takeway Motivation
▶
Clients prefer streaming YouTube videos over IPv6.
▶
Observed performance (both in terms of latency and throughput) over IPv6 is worse.
▶
Stall rates are low, but when a stall occurs, stall durations over IPv6 are higher.
▶
Worse performance due to GGC nodes that are IPv4-only.
Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Throughput Stall Events
▶
Reproducibility Considerations: ▶
The test is open-sourced: https://github.com/sabyahsan/youtube-test
▶
The dataset is released: https://github.com/vbajpai/2017-ccr-youtube-analysis
Recommendations Q/A
www.vaibhavbajpai.com
[email protected] | @bajpaivaibhav 18 / 18
References [1] NANOG. (2016) IPv6 traffic percentages? [Online]. Available: https: //mailman.nanog.org/pipermail/nanog/2016-January/083624.html
Standard), RFC Editor, Fremont, CA, USA, pp. 1–32, Sep. 2012. [Online]. Available: https://www.rfc-editor.org/rfc/rfc6724.txt
[2] N. Sarrar, G. Maier, B. Ager, R. Sommer, and S. Uhlig, “Investigating IPv6 Traffic - What Happened at the World IPv6 Day?” ser. Passive and Active Measurement Conference, 2012, pp. 11–20. [Online]. Available: https://doi.org/10.1007/978-3-642-28537-0_2
[5] D. Wing and A. Yourtchenko, “Happy Eyeballs: Success with Dual-Stack Hosts,” RFC 6555 (Proposed Standard), RFC Editor, Fremont, CA, USA, pp. 1–15, Apr. 2012. [Online]. Available: https://www.rfc-editor.org/rfc/rfc6555.txt
[3] G. Developers. (2017) YouTube Data API. [Online]. Available: https://developers.google.com/youtube/v3/docs/videos/list
[6] V. Bajpai and J. Schönwälder, “Measuring the Effects of Happy Eyeballs,” ser. Applied Networking Research Workshop, 2016. [Online]. Available: http://dl.acm.org/citation.cfm?id=2959429
[4] D. Thaler, R. Draves, A. Matsumoto, and T. Chown, “Default Address Selection for Internet Protocol Version 6 (IPv6),” RFC 6724 (Proposed
Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Throughput Stall Events Recommendations Q/A
18 / 18