using Large Scale Measurement Platforms Understanding the Impact of - PowerPoint PPT Presentation

Contributions IM 2017 Conference May 2017 University of Ghent, Belgium Filip De Turck Jacobs University Bremen, Germany Kinga Lipskoch Jacobs University Bremen, Germany Jürgen Schönwälder Dissertation Committee Lisbon, Portugal TU Munich IPv6 Performance Vaibhav Bajpai using Large Scale Measurement Platforms Understanding the Impact of Network Infrastructure Changes Q/A Tianks! Last-mile Latency Happy Eyeballs YouTube Latency Failures 1 / 19

Contributions Q/A * entries are papers currently under review. Measuring Access Network Performance IPv6 Performance Measuring IPv6 Performance Survey on Internet Performance Measurement Platforms [1] 2 / 19 Tianks! Last-mile Latency Happy Eyeballs YouTube Failures Latency [ COMST ′ 15 ] [ CNSM ′ 16 ] ▶ Measuring Web Similarity [2] [ NETWORKING ′ 15 ] ▶ Measuring TCP Connect Times [3] ▶ Measuring YouTube Performance [4] [ PAM ′ 15 ] ▶ Measuring Efgects of Happy Eyeballs [5] [ ANRW ′ 16 ] ▶ RIPE Atlas Vantage Point Selection [6] [ IM ′ 17 ] ▶ Dissecting Last-mile Latency Characteristics [ ∗ ] ▶ Lessons Learned from using RIPE Atlas [7] [ SIGCOMM CCR ′ 15 ]

Contributions Tianks! IPv6 Performance IPv6 Performance Q/A shaded region represents the duration of the longitudinal study. Last-mile Latency Failures Happy Eyeballs 3 / 19 YouTube Latency Google IPv6 Adoption 15% ▶ Literature focus largely on IPv6 adoption. 10% 5% ▶ Very little work on measuring IPv6 performance. 0% 2009 2010 2011 2012 2013 2014 2015 2016 2017 ▶ Tiis study closes the gap. 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 We measure from ∼ 100 dual-stacked SamKnows probes.

Contributions Tianks! Metrics should account for changes in IPv6-readiness. IPv6 Performance Complete Failures Q/A 4 / 19 Last-mile Latency Latency YouTube Failures Happy Eyeballs www.bing.com 10 3 ALEXA 1M with AAAA entries 10 2 50.0% 10 1 40.0% W6D W6LD 10 0 HTTP Failure 30.0% www.detik.com 10 3 20.0% 10.0% 0.0% 10 2 2010 2011 2012 2013 2014 2015 2016 2017 www.engadget.com TCP Connect Times (ms) 10 3 10 2 10 1 ▶ Failures reduced from 40% (2009) to 3% today. 10 0 www.nifty.com 10 3 10 2 Failing AAAA Websites www.qq.com 1.0 10 4 [Mar '17] 0.8 10 3 10 2 0.6 CDF 10 1 0.4 10 0 4.3K 0.2 www.sakura.ne.jp 0.0 10 3 0.1K 1K 10K 100K 1000K ALEXA Rank 10 2 Jul Jul Jul Jan Jan Jan Jan 2013 2014 2015 2016 IPv6 IPv4 ▶ 88% failing websites rank > 100K. ▶ 1% rank < 10K, six websites rank < 300.

Contributions overestimation of IPv6 adoption numbers IPv6 Performance IPv4 W6LD Success Rate (%) Webpage # tools that identify such partial failures failures can be deemed IPv6-ready 5 / 19 Failures Latency YouTube Happy Eyeballs Last-mile Latency Tianks! Q/A Partial Failures IPv6 ( ↓ ) ALEXA top 100 websites with AAAA entries. 01 www.bing.com 0 100 ✓ ▶ 27% show some rate of failure over IPv6. 02 www.detik.com 0 100 ✓ 03 www.engadget.com 0 100 ✓ 04 www.nifty.com 0 100 ▶ 9% exhibit more than 50% failures over IPv6. 05 www.qq.com 0 100 06 www.sakura.ne.jp 0 100 07 www.flipkart.com 09 99 ✓ 08 www.folha.uol.com.br 13 100 1.0 IPv6 (100) 0.8 09 www.aol.com 48 100 ✓ 0.6 IPv4 (100) CDF 0.4 10 www.comcast.net 52 100 ✓ 0.2 11 www.yahoo.com 72 100 ✓ 0.0 12 www.mozilla.org 84 100 ✓ 0 20 40 60 80 100 13 www.orange.fr 86 100 ✓ Success Rate (%) 14 www.seznam.cz 89 100 ✓ 15 www.mobile.de 90 100 ✓ 16 www.wikimedia.org 90 100 ▶ Limiting to root webpage can lead to 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 ✓ ▶ Unclear whether websites with partial 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 ▶ ISOC now supporting [8] development of ✓ 25 www.marca.com 97 100 ✓ 26 www.terra.com.br 98 100 ✓ 27 www.youm7.com 99 100

Contributions Q/A networks relevant for upcoming IPv6-only notice them due to IPv4 fallback. IPv6 Performance Partial Failures | Root Cause Analysis 6 / 19 Tianks! Happy Eyeballs Failures Latency YouTube Last-mile Latency Website failing over IPv6 Network Level Content Level Service Level www.bing.com (100%) CURLE_OK www.detik.com (100%) CURLE_COULDNT_RESOLVE_HOST www.engadget.com (100%) CURLE_COULDNT_CONNECT www.nifty.com (100%) CURLE_OPERATION_TIMEDOUT www.qq.com (100%) CURLE_GOT_NOTHING www.sakura.ne.jp (100%) CURLE_RECV_ERROR www.flipkart.com (91%) ▶ Failures silently exist; clients do not www.folha.uol.com.br (87%) www.aol.com (52%) */css www.comcast.net (48%) */html www.yahoo.com (28%) */javascript, */json www.mozilla.org (16%) */octet-stream www.orange.fr (14%) */plain ▶ Identifjcation of operational issues www.seznam.cz (11%) */rdf www.mobile.de (10%) */xml www.wikimedia.org (10%) image/* www.t-online.de (7%) www.free.fr (5%) www.usps.com (5%) SAME ORIGIN www.vk.com (5%) CROSS ORIGIN www.wikipedia.org (5%) www.wiktionary.org (5%) www.elmundo.es (4%) www.uol.com.br (4%) www.marca.com (3%) www.terra.com.br (2%) www.youm7.com (1%) 0 30 60 90 0 30 60 90 0 30 60 90 Contribution (%) ▶ Failures due to DNS resolution error on image/*, */javascript, */json and */css content. ▶ 12% websites have more than 50% content that belongs to same-origin source and fails over IPv6, ▶ Content failing from cross-origin sources consists of analytics and third-party advertisements.

Contributions Tianks! IPv6 Performance ensure their CDN caches are dual-stacked. Latency | Websites Q/A 7 / 19 Last-mile Latency Failures YouTube Happy Eyeballs Latency TCP Connect Times [∆sa (ms)] 50 0 −50 −100 ∆ s a ( u ) = t 4 ( u ) − t 6 ( u ) −150 where t ( u ) is the time taken to establish TCP connection to website u . www.bing.com www.wikipedia.org www.facebook.com www.youtube.com 20 0 ▶ ISPs in early stages of IPv6 deployment should −20 −40 −60 2013 2014 2015 2016 2017 www.blogspot.* www.netflix.com www.google.* 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

Contributions Tianks! IPv6 Performance ALEXA top 10K websites (as of Jan 2017): Latency | Websites - Who connects faster? Q/A 8 / 19 Last-mile Latency Failures YouTube Latency Happy Eyeballs 1.0 openstreetmap 0.8 cloudflare [01/2017] heise ▶ 40% are faster over IPv6. 0.6 google wikipedia CDF facebook 0.4 microsoft ▶ 94% of the rest are at most 1 ms slower. yahoo linkedin 0.2 netflix ALEXA (10K) ▶ 3% are at least 10 ms slower. 0.0 −1.0 −0.5 0.0 0.5 1.0 ▶ 1% are at least 100 ms slower. ∆sa (ms) ∆ s a ( u ) = t 4 ( u ) − t 6 ( u ) ▶ Relevant for content providers to get insights on how their service delivery compares over IPv6.

Contributions Tianks! IPv6 Performance Latency is consistently higher over IPv6. YouTube Q/A 9 / 19 Last-mile Latency Happy Eyeballs Failures Latency YouTube TCP Connect Times 0 ∆t (ms) −1 −2 −3 Web −4 −5 ▶ TCP connect times TCP Connect Times 0.0 ∆t (ms) −0.1 Audio ▶ < 1 ms slower over IPv6 −0.2 Video ▶ Higher towards webpages −0.3 −0.4 Prebuffering Duration ▶ Prebufgering durations ∆p (ms) 0 −40 −80 ▶ > 25 ms slower over IPv6 −120 Startup Delay 0 ▶ Startup delay ∆s (ms) −100 −200 −300 ▶ > 100 ms slower over IPv6 −400 Oct Jan Apr Jul Oct Jan Apr 2015 2016 ∆ t ( y ) = tc 4 ( y ) − tc 6 ( y ) ▶ ISPs should make their GGC nodes dual-stacked. ∆ p ( y ) = pd 4 ( y ) − pd 6 ( y ) ∆ s ( y ) = sd 4 ( y ) − sd 6 ( y )

Contributions Q/A IPv6 Performance same IPv6 preference levels. HE timer value > 300 ms Happy Eyeballs 10 / 19 Tianks! Latency Happy Eyeballs Last-mile Latency YouTube Failures 1.0 ['13 - '17] IPv6 (462K) 0.8 300 ms ▶ Only ∼ 1% of samples above CDF 0.6 IPv4 (462K) 0.4 0.2 0.0 10 -2 10 -1 10 0 10 1 10 2 10 3 10 4 TCP Connect Times (ms) Samples where HE prefers IPv6 − ▶ HE prefers slower IPv6 1.0 93% ['13 - '17] 462K 0.8 connections 90% of the time. 99% 0.6 CDF 30% ▶ HE timer of 150 ms maintains 0.4 1% 7% 2% 0.2 0.0 −40 −30 −20 −10 0 10 ∆sa (ms) ▶ RFC 6555 should have used 150 ms timer. Measurements should inform protocol engineering. ▶ Drive an RFC 6555 update with operational experience within the IETF.

Contributions Q/A * entries are papers currently under review. Measuring Access Network Performance IPv6 Performance Measuring IPv6 Performance Survey on Internet Performance Measurement Platforms 11 / 19 Tianks! Last-mile Latency Happy Eyeballs YouTube Failures Latency [ COMST ′ 15 ] ▶ Measuring Web Similarity [ CNSM ′ 16 ] ▶ Measuring TCP Connect Times [ NETWORKING ′ 15 ] ▶ Measuring YouTube Performance [ PAM ′ 15 ] ▶ Measuring Efgects of Happy Eyeballs [ ANRW ′ 16 ] ▶ RIPE Atlas Vantage Point Selection [ IM ′ 17 ] ▶ Dissecting Last-mile Latency Characteristics [ ∗ ] ▶ Lessons Learned from using RIPE Atlas [ SIGCOMM CCR ′ 15 ]