Failures, Latency and Happy Eyeballs Q/A Recommendations YouTube - - PowerPoint PPT Presentation

▶

Feb 28, 2023 149 likes •497 views

Websites Joint work with Vaibhav Bajpai TU Munich CAIDA San Diego, USA Under revision at: IEEE/ACM Transactions on Networking Jrgen Schnwlder Motivation Jacobs University Bremen, Germany Steffje Jacob Eravuchira Samknows Limited,

SLIDE 1

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

A Longitudinal View of Dual-stacked Websites − Failures, Latency and Happy Eyeballs

Vaibhav Bajpai TU Munich CAIDA San Diego, USA

Under revision at: IEEE/ACM Transactions on Networking

Joint work with Jürgen Schönwälder Jacobs University Bremen, Germany Steffje Jacob Eravuchira Samknows Limited, London, UK Sam Crawford Samknows Limited, London, UK August 2017

1 / 33

SLIDE 2

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Motivation

▶ Literature focus largely on IPv6 adoption. ▶ Very little work on measuring IPv6 performance. ▶ Tiis study closes the gap.

2009 2010 2011 2012 2013 2014 2015 2016 2017 0% 5% 10% 15% Google IPv6 Adoption shaded region represents the duration of the longitudinal study.

∼100 dual-stacked SamKnows probes (∼66 difgerent origin ASes)

NETWORK TYPE # RESIDENTIAL 78 NREN / RESEARCH 10 BUSINESS / DATACENTER 08 OPERATOR LAB 04 IXP 01 RIR # RIPE 60 ARIN 29 APNIC 10 AFRINIC 01 LACNIC 01 2 / 33

SLIDE 3

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Motivation

1.0%

0.0% 1.0% 2.0% 3.0%

Google IPv6 Adoption

Native IPv6 2008 2009 2010 2011 2012 2013 2014 2015 2016

0.02%

0.0% 0.02% 0.04% 0.06% Weekends - Weekdays Teredo/6to4

▶ Tie drifu of IPv6 penetration is increasing.

Country Distribution

Japan: 29.63 % Japan: 29.63 % Peru: 20.37 % Peru: 20.37 % Malaysia: 3.70 % Malaysia: 3.70 % United States: 1.85 % United States: 1.85 % Brazil: 16.67 % Brazil: 16.67 % Pakistan: 1.85 % Pakistan: 1.85 % Argentina: 3.70 % Argentina: 3.70 % Kenya: 1.85 % Kenya: 1.85 % Australia: 3.70 % Australia: 3.70 % South Africa: 1.85 % South Africa: 1.85 % Sri Lanka: 3.70 % Sri Lanka: 3.70 % Bulgaria: 1.85 % Bulgaria: 1.85 % Zambia: 1.85 % Zambia: 1.85 % Taiwan: 1.85 % Taiwan: 1.85 % Botswana: 1.85 % Botswana: 1.85 % Canada: 3.70 % Canada: 3.70 % Highcharts.com

▶ Prefjxes blacklisted by Google over IPv6 (2017)

2010 2011 2012 2013 2014 2015 2016 2017 0.0% 5.0% 10.0% 15.0% 20.0% W6D W6LD ALEXA 1M with AAAA entries 0% 10% 20% 30% 40% Websites top 0.1K top 1K top 10K top 100K top 1000K 40.0 25.7 20.4 17.4 12.1

▶ Cloudfmare Efgect (shaded region) ▶ Tie dent created by Cloudfmare > W6D (or W6LD). ▶ Cloudfmare added AAAA entries for all websites [1]. A CDN plays a leading role in technology adoption and shifuing signifjcant traffjc overnight over IPv6.

3 / 33

SLIDE 4

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Research Contribution

▶ Complete Failures ▶ Partial Failures

▶ Failures silently exist; clients do not notice them due to IPv4 fallback. ▶ Can websites with partial failures be deemed IPv6-ready? ▶ Quantifjcation of failures helpful for upcoming IPv6-only networks.

▶ Latency ▶ Happy Eyeballs

Tiis is the fjrst study to provide a longitudinal view (4 years)

f failures and performance of dual-stacked websites.

4 / 33

SLIDE 5

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Complete Failures

2010 2011 2012 2013 2014 2015 2016 2017 0.0% 10.0% 20.0% 30.0% 40.0% 50.0% W6D W6LD ALEXA 1M with AAAA entries HTTP Failure

▶ Failures reduced from 40% (2009) to 3% today.

0.1K 1K 10K 100K 1000K ALEXA Rank 0.0 0.2 0.4 0.6 0.8 1.0 CDF Failing AAAA Websites

4.3K

[Mar '17]

▶ 88% failing websites rank > 100K. ▶ 1% rank < 10K, six websites rank < 300.

100 101 102 103 www.bing.com 102 103 www.detik.com 100 101 102 103 www.engadget.com 102 103 www.nifty.com 100 101 102 103 104 www.qq.com Jan 2013 Jan 2014 Jan 2015 Jan 2016 Jul Jul Jul 102 103 www.sakura.ne.jp IPv6 IPv4 TCP Connect Times (ms)

Metrics should account for changes in IPv6-readiness.

5 / 33

SLIDE 6

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Complete Failures | WL6D websites

▶ ∼3K websites participated on the W6LD (2012). ▶ Promise to permanently enable production-ready IPv6 on the Internet.

W6LD websites −

▶ 8% (259) do not have A or AAAA entries in DNS.

W6LD websites (with A and AAAA entries) −

▶ 1% (36) TCP timeout over both AF. ▶ 8% (253) TCP timeout over IPv6.

1K 10K 100K 1M 10M 100M ALEXA Rank 0.0 0.2 0.4 0.6 0.8 1.0 CDF Failing W6LD Websites

253

▶ 3% failing websites rank < 10K. ▶ 75% rank > 100K, 61% rank > 1M.

6 / 33

SLIDE 7

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Partial Failures

ALEXA top 100 websites with AAAA entries. ▶ 27% show some rate of failure over IPv6. ▶ 9% exhibit more than 50% failures over IPv6.

20 40 60 80 100 Success Rate (%) 0.0 0.2 0.4 0.6 0.8 1.0 CDF IPv6 (100) IPv4 (100)

▶ Limiting to root webpage can lead to

verestimation of IPv6 adoption numbers

▶ Unclear whether websites with partial failures can be deemed IPv6-ready ▶ ISOC now supporting [2] development of tools that identify such partial failures

# Webpage Success Rate (%) W6LD IPv6(↓) IPv4 01 www.bing.com 100 ✓ 02 www.detik.com 100 ✓ 03 www.engadget.com 100 ✓ 04 www.nifty.com 100 05 www.qq.com 100 06 www.sakura.ne.jp 100 07 www.flipkart.com 09 99 ✓ 08 www.folha.uol.com.br 13 100 09 www.aol.com 48 100 ✓ 10 www.comcast.net 52 100 ✓ 11 www.yahoo.com 72 100 ✓ 12 www.mozilla.org 84 100 ✓ 13 www.orange.fr 86 100 ✓ 14 www.seznam.cz 89 100 ✓ 15 www.mobile.de 90 100 ✓ 16 www.wikimedia.org 90 100 17 www.t-online.de 93 100 ✓ 18 www.free.fr 95 100 19 www.usps.com 95 100 20 www.vk.com 95 100 ✓ 21 www.wikipedia.org 95 100 ✓ 22 www.wiktionary.org 95 100 23 www.elmundo.es 96 100 ✓ 24 www.uol.com.br 96 100 ✓ 25 www.marca.com 97 100 ✓ 26 www.terra.com.br 98 100 ✓ 27 www.youm7.com 99 100 7 / 33

SLIDE 8

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Partial Failures | Root Cause Analysis

30 60 90 www.youm7.com (1%) www.terra.com.br (2%) www.marca.com (3%) www.uol.com.br (4%) www.elmundo.es (4%) www.wiktionary.org (5%) www.wikipedia.org (5%) www.vk.com (5%) www.usps.com (5%) www.free.fr (5%) www.t-online.de (7%) www.wikimedia.org (10%) www.mobile.de (10%) www.seznam.cz (11%) www.orange.fr (14%) www.mozilla.org (16%) www.yahoo.com (28%) www.comcast.net (48%) www.aol.com (52%) www.folha.uol.com.br (87%) www.flipkart.com (91%) www.sakura.ne.jp (100%) www.qq.com (100%) www.nifty.com (100%) www.engadget.com (100%) www.detik.com (100%) www.bing.com (100%) Network Level

CURLE_OK CURLE_COULDNT_RESOLVE_HOST CURLE_COULDNT_CONNECT CURLE_OPERATION_TIMEDOUT CURLE_GOT_NOTHING CURLE_RECV_ERROR

30 60 90 Contribution (%) Content Level

*/css */html */javascript, */json */octet-stream */plain */rdf */xml image/*

30 60 90 Service Level

SAME ORIGIN CROSS ORIGIN

Website failing over IPv6

▶ Failures due to DNS resolution error on

image/*, */javascript, */json and */css content.

▶ Failures silently exist; clients do not notice them due to IPv4 fallback. ▶ Identifjcation of operational issues relevant for upcoming IPv6-only networks

8 / 33

SLIDE 9

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Partial Failures | Root Cause Analysis

30 60 90 Contribution (%) www.youm7.com (1%) www.terra.com.br (2%) www.marca.com (3%) www.uol.com.br (4%) www.elmundo.es (4%) www.wiktionary.org (5%) www.wikipedia.org (5%) www.vk.com (5%) www.usps.com (5%) www.free.fr (5%) www.t-online.de (7%) www.wikimedia.org (10%) www.mobile.de (10%) www.seznam.cz (11%) www.orange.fr (14%) www.mozilla.org (16%) www.yahoo.com (28%) www.comcast.net (48%) www.aol.com (52%) www.folha.uol.com.br (87%) www.flipkart.com (91%) www.sakura.ne.jp (100%) www.qq.com (100%) www.nifty.com (100%) www.engadget.com (100%) www.detik.com (100%) www.bing.com (100%) *.youm7.com *.terra.com.br *.marca.com *.uol.com.br *.elmundo.es *.wiktionary.org *.wikipedia.org *.vk.com *.usps.com *.free.fr *.t-online.de *.wikimedia.org *.mobile.de *.seznam.cz *.orange.fr *.mozilla.org *.yahoo.com *.comcast.net *.aol.com *.uol.com.br *.flipkart.com *.sakura.ne.jp *.qq.com *.nifty.com *.engadget.com *.detik.com *.bing.com SAME ORIGIN

▶ 12% of websites have more than 50% webpage elements that belong to the same origin source and fail over IPv6. ▶ CDN infrastructure does not have IPv6 turned on by default for all same-origin webpage elements.

# Webpage Same Origin (↓) 01 www.bing.com 100% 02 www.detik.com 100% 03 www.engadget.com 100% 04 www.nifty.com 100% 05 www.usps.com 100% 06 www.qq.com 100% 07 www.sakura.ne.jp 100% 08 www.comcast.net 85% 09 www.yahoo.com 83% 10 www.terra.com.br 74% 11 www.marca.com 70% 12 www.wikimedia.org 65% 13 www.elmundo.es 37% 14 www.vk.com 31% 15 www.t-online.de 30% 16 www.youm7.com 24% 17 www.wiktionary.org 22% 18 www.wikipedia.org 22% 19 www.free.fr 13% 20 www.folha.uol.com.br 12% 21 www.mozilla.org 7% 22 www.uol.com.br 7% 23 www.mobile.de 7% 24 www.aol.com 5% 25 www.orange.fr 5% 26 www.seznam.cz 4% 27 www.flipkart.com 1% 9 / 33

SLIDE 10

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Partial Failures | Root Cause Analysis

*.adition.com *.ajax.googleapis.com *.aolcdn.com *.cimcontent.net *.creativecommons.org *.d5nxst8fruw4z.cloudfront.net *.demdex.net *.dmtry.com *.doubleclick.net *.el-mundo.net *.elmundo.es *.expansion.com *.f.i.uol.com.br *.flixcart.com *.globaliza.com *.images1.folha.com.br *.imedia.cz *.imguol.com *.imguol.com.br *.interactivemedia.net *.ioam.de *.jsuol.com.br *.leguide.com *.ligatus.com *.mail.ru *.mozilla.net *.navdmp.com *.netbiscuits.net *.omtrdc.net *.optimizely.com *.outbrain.com *.proxad.net *.quantserve.com *.sblog.cz *.scorecardresearch.com *.szn.cz *.tag.navdmp.com *.telva.com *.theadex.com *.toi.de *.trrsf.com *.unidadeditorial.es *.voila.fr *.woopic.com *.www1.folha.com.br *.xiti.com

▶ Tiird-party advertisements (.doubleclick.net) ▶ Analytics (.scorecardresearch.com, .quantserve.com) ▶ User-centric content (.facebook.com, .ajax.googleapis.com) ▶ Static content (.wikimedia.org, *.creativecommons.org) ▶ Enabling IPv6 on few cross cross-origin sources (creativecommons.org, doubleclick.net) will help reduce partial failure of multiple websites.

10 / 33

SLIDE 11

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Latency | Websites

∆sa(u) = t4(u) − t6(u) where t(u) is the time taken to establish TCP connection to website u.

▶ ISPs in early stages of IPv6 deployment should ensure their CDN caches are dual-stacked.

−150 −100 −50 50 TCP Connect Times [∆sa (ms)] www.bing.com www.facebook.com www.wikipedia.org www.youtube.com 2013 2014 2015 2016 2017 −60 −40 −20 20 www.blogspot.* www.google.* www.netflix.com www.yahoo.com

▶ TCP connect times to popular websites over IPv6 have considerably improved over time. ▶ Infmated latency over IPv6 was due to missing content caches over IPv6

11 / 33

SLIDE 12

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Latency | Websites - Who connects faster?

ALEXA top 10K websites (as of Jan 2017):

▶ 40% are faster over IPv6. ▶ 94% of the rest are at most 1 ms slower. ▶ 3% are at least 10 ms slower. ▶ 1% are at least 100 ms slower.

−1.0 −0.5 0.0 0.5 1.0 ∆sa (ms) 0.0 0.2 0.4 0.6 0.8 1.0 CDF netflix yahoo google linkedin microsoft facebook wikipedia cloudflare heise

penstreetmap

ALEXA (10K) [01/2017] ∆sa(u) = t4(u) − t6(u)

▶ Relevant for content providers to get insights on how their service delivery compares over IPv6.

12 / 33

SLIDE 13

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Happy Eyeballs

▶ Only ∼1% of samples above

HE timer value > 300 ms

Websites where HE prefers IPv6 −

▶ A 300 ms HE timer value leaves

2% chance for IPv4.

▶ 99% of top 10K ALEXA prefer

IPv6 98% of time.

10-2 10-1 100 101 102 103 104 TCP Connect Times (ms) 0.0 0.2 0.4 0.6 0.8 1.0

CDF

300 ms IPv6 (462K) IPv4 (462K) ['13 - '17] 96% 97% 98% 99% 100% 0.0 0.2 0.4 0.6 0.8 1.0 CCDF PROBES (80) ALEXA (10K) ['13 - '17] Preference (300 ms) 13 / 33

SLIDE 14

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Happy Eyeballs

Samples where HE prefers IPv6 −

▶ HE prefers slower IPv6

connections 90% of the time.

▶ HE timer of 150 ms maintains

same IPv6 preference levels.

▶ We get margin benefjt of 10%

because timer cuts early.

−40 −30 −20 −10 10 ∆sa (ms) 0.0 0.2 0.4 0.6 0.8 1.0 CDF 1% 2% 7% 30% 93% 99% 462K ['13 - '17] 50 100 150 200 250 300 HE timer (ms) 0% 20% 40% 60% 80% 100% Preference 150 ms

ALEXA (10K) ['13 - '17]

▶ RFC 6555 should have used 150 ms timer. Measurements should inform protocol engineering. ▶ Drive an RFC 6555 update with operational experience within the IETF.

14 / 33

SLIDE 15

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Impact

▶ A Longitudinal View of Dual-Stacked Websites −

▶ Failures [3] [CNSM ′16] ▶ Latency [4] and [NETWORKING ′15] ▶ Happy Eyeballs [5] [ANRW ′16]

▶ Relevance:

▶ Network operators in early stages of IPv6 deployment. ▶ Content providers to see how their service delivery over IPv6 compares to IPv4. ▶ Drive related standards work in the IETF. 15 / 33

SLIDE 16

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Measuring YouTube Content Delivery over IPv6

Vaibhav Bajpai TU Munich CAIDA San Diego, USA

Published at: SIGCOMM Computer Communication Review October 2017: https://goo.gl/oUJ7Ej

Joint work with Saba Ahsan Aalto University, Helsinki, Finland Jörg Ott TU Munich, Germany Jürgen Schönwälder Jacobs University Bremen, Germany August 2017

16 / 33

SLIDE 17

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Motivation

▶ IPv6 contributes ∼25% [6] of traffjc within Comcast. ▶ Swisscom reports ∼60% [6] of IPv6 traffjc is YouTube. ▶ IPv6 traffjc largely dominated by YouTube [7].

2009 2010 2011 2012 2013 2014 2015 2016 2017 0% 5% 10% 15% Google IPv6 Adoption

shaded region represents the duration of the longitudinal study.

Do users experience benefjt (or sufger) from YouTube streaming over IPv6? ∼100 dual-stacked SamKnows probes (∼66 difgerent origin ASes)

NETWORK TYPE # RESIDENTIAL 78 NREN / RESEARCH 10 BUSINESS / DATACENTER 08 OPERATOR LAB 04 IXP 01 RIR # RIPE 60 ARIN 29 APNIC 10 AFRINIC 01 LACNIC 01 17 / 33

SLIDE 18

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Research Contribution

▶ 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.

Tiis is the fjrst study to measure YouTube content delivery over IPv6

18 / 33

SLIDE 19

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Methodology | Selection of YouTube Videos

▶ Using YouTube v3 API [8]. ▶ Video Selection Criteria:

1. Video duration > 60s.
2. Available in Full HD.
3. No regional restrictions.

▶ List updated every 12h. ▶ Probes daily pull the list. ▶ Tie test supports non-adaptive and step-down playout modes only. ▶ Results are biased our vantage points (centered largely around EU, US and JP).

19 / 33

SLIDE 20

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Success Rate

▶ Number of successful

iterations to total iterations.

▶ Tie test executes once every

hour (over both AF).

0.0 0.2 0.4 0.6 0.8 1.0 CCDF Success Rate IPv6 (102) 80% 85% 90% 95% 100% 0.0 0.2 0.4 0.6 0.8 1.0 IPv4 (102) ['14 - '17]

▶ 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.

20 / 33

SLIDE 21

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

IPv6 Preference

95% 96% 97% 98% 99% 100% 0.0 0.2 0.4 0.6 0.8 1.0 CCDF Web (458) Audio (458) Video (458)

['14 - '16]

IPv6 Preference

▶ RFC 6724 [?] makes apps prefer connections made over IPv6. ▶ RFC 6555 [?] allows apps to fallback to IPv4 when IPv6 connectivity is bad. ▶ TCP connections over IPv6 are preferred at least 97% of the time.

Clients prefer streaming YouTube videos over IPv6

21 / 33

SLIDE 22

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

TCP connect times

−100 −75 −50 −25 25 50 ∆t (ms) 0.0 0.2 0.4 0.6 0.8 1.0 CDF Web (6.5M) Audio (6.5M) Video (6.5M) ['14 - '17] TCP connect times

▶ 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.

22 / 33

SLIDE 23

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

TCP connect times

−5 ∆t (ms) TCP Connect Times Web Jan 2015 Jan 2016 Jan 2017 Jul Jul Jul −0.4 −0.3 −0.2 −0.1 0.0 ∆t (ms) TCP Connect Times Audio Video

▶ 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.

23 / 33

SLIDE 24

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Sequence Diagram (contd.)

24 / 33

SLIDE 25

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Startup Delay

−5 −4 −3 −2 −1 1 ∆s (s) 0.0 0.2 0.4 0.6 0.8 1.0 CDF 6.5M ['14 - '17] Startup Delay

▶ 80% of the samples are slower over IPv6. ▶ Half of the samples are at least 100 ms slower over IPv6.

25 / 33

SLIDE 26

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Startup Delay

▶ Prebufgering 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.

26 / 33

SLIDE 27

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Sequence Diagram (contd.)

27 / 33

SLIDE 28

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Tiroughput

−4 −2 2 4 ∆tp (MB/s) 0.0 0.2 0.4 0.6 0.8 1.0 CDF

Video (6.5M) Audio (6.5M)

['14 - '17] Throughput

Jan 2015 Jan 2016 Jan 2017 Jul Jul Jul −1.0 −0.7 −0.4 −0.1 0.2 ∆tp (MB/s) Throughput Video Audio

▶ 80% of video and 60% audio samples achieve lower throughput over IPv6. ▶ Tie throughput is consistently lower over IPv6, but it has improved over time.

28 / 33

SLIDE 29

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Sequence Diagram (contd.)

29 / 33

SLIDE 30

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Stall Rates

0.0 0.2 0.4 0.6 0.8 1.0 CDF Stall Rate IPv6 (102) 0% 5% 10% 15% 20% 25% 30% 0.0 0.2 0.4 0.6 0.8 1.0 IPv4 (102) ['14 - '17]

▶ 90% of the probes witness less than 1% stall rate over both address families. ▶ Bitrates reliably streamed is also comparable over both address families.

30 / 33

SLIDE 31

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Stall Durations

−60 −40 −20 20 40 60 ∆st (s) 0.0 0.2 0.4 0.6 0.8 1.0 CDF 1.7K ['14 - '17] Stall Durations

▶ 80% of samples experience stall durations that are at least 1s longer.

31 / 33

SLIDE 32

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Recommendations

▶ Update RFC 6555 with a lower HE timer value.

▶ We have shown [5] that reducing HE timer value to 150 ms (from 300 ms) helps.

2009 2010 2011 2012 2013 2014 2015 2016 2017 0.00% 0.05% 0.10% 0.15% 0.20% W6D W6LD 6to4/Teredo Decline

▶ ISPs should put latency as a fjrst-class citizen. ▶ ISPs should ensure GGC nodes are dual-stacked.

▶ Request an IPv6 prefjx allocation from Google. 32 / 33

SLIDE 33

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

Takeway

▶ 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. ▶ Reproducibility Considerations:

▶ Tie test is open-sourced: https://github.com/sabyahsan/youtube-test ▶ Tie dataset is released: https://github.com/vbajpai/2017-ccr-youtube-analysis

www.vaibhavbajpai.com bajpaiv@in.tum.de | @bajpaivaibhav

33 / 33

SLIDE 34

Websites

Motivation Contribution Complete Failures Partial Failures Latency Happy Eyeballs

Q/A YouTube

Motivation Methodology Success Rate IPv6 Preference TCP connect times Startup Delay Tiroughput Stall Events Recommendations

Q/A

References

[1] “98.01% of sites on Cloudfmare now use IPv6,” https://blog.cloudflare.com/98-percent-ipv6, [Online; accessed 15-Apr-2017]. [2] “NAT64 Check,” nat64check.ipv6-lab.net, [Accessed 15-Apr-2017]. [3]

S. J. Eravuchira, V. Bajpai, J. Schönwälder, and S. Crawford,

“Measuring Web Similarity from Dual-stacked Hosts,” ser. Conference on Network and Service Management, 2016, pp. 181–187. [Online]. Available: http://dx.doi.org/10.1109/CNSM.2016.7818415 [4]

V. Bajpai and J. Schönwälder, “IPv4 versus IPv6 - who connects

faster?” ser. IFIP Networking Conference, 2015, pp. 1–9. [Online]. Available: http://dx.doi.org/10.1109/IFIPNetworking.2015.7145323 [5] ——, “Measuring the Efgects of Happy Eyeballs,” ser. Applied Networking Research Workshop, 2016. [Online]. Available: http://dl.acm.org/citation.cfm?id=2959429 [6] “NANOG - IPv6 traffjc percentages?” http://goo.gl/w1mJ8y, [Online; accessed 05-May-2016]. [7]

N. Sarrar, G. Maier, B. Ager, R. Sommer, and S. Uhlig, “Investigating

ipv6 traffjc,” in Passive and Active Measurement, ser. Lecture Notes in Computer Science, N. Tafu and F. Ricciato, Eds. Springer Berlin Heidelberg, 2012, vol. 7192, pp. 11–20. [Online]. Available: http://dx.doi.org/10.1007/978-3-642-28537-0_2 [8] “YouTube Data API,” https://goo.gl/EBCbR8, [Online; accessed 27-Apr-2016]. 33 / 33