Vantage Point Selection for IPv6 Measurements Takeway User Tags - - PowerPoint PPT Presentation

▶

Apr 11, 2024 138 likes •308 views

Introduction Lisbon, Portugal May 2017 RIPE NCC, Amsterdam Emile Aben RIPE NCC, Amsterdam Robert Kisteleki Jacobs University, Bremen Jrgen Schnwlder SamKnows Limited, London Steffje Jacob Eravuchira Joint work with IM 2017

SLIDE 1

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

Vantage Point Selection for IPv6 Measurements

Benefjts and Limitations of RIPE Atlas Tags

Vaibhav Bajpai

TU Munich

IM 2017 Conference Lisbon, Portugal

Joint work with Steffje Jacob Eravuchira SamKnows Limited, London Jürgen Schönwälder Jacobs University, Bremen Robert Kisteleki RIPE NCC, Amsterdam Emile Aben RIPE NCC, Amsterdam May 2017

1 / 15

SLIDE 2

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

Introduction | RIPE Atlas [1, 2]

▶ A platform of hardware-based probes that run active measurements on the Internet. ▶ Plays critical role in −

▶ providing operational support to network operators. ▶ facilitating measurement-based research.

▶ ∼9.6K probes connected around the globe (May 2017).

2010 2011 2012 2013 2014 2015 2016 2017 5K 10K 15K 20K API STATUS RIPE Atlas Probes NEVER CONNECTED CONNECTED DISCONNECTED ABANDONED REGISTERED

NEVER CONNECTED 2965 13.48% CONNECTED 9685 44.04% DISCONNECTED 2009 09.14% ABANDONED 7334 33.34% REGISTERED 21993 100.0% 2 / 15

SLIDE 3

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

Introduction | RIPE Atlas Measurements

Feb ′13 Public APIs [3, 4, 5] to provision measurements on probes.

MEASUREMENT TARGET ping, ping6 first hop, second hop (derived from tracer-

ute measurements), *.root-servers.net,

*.atlas.ripe.net traceroute, traceroute6 *.root-servers.net, *.atlas.ripe.net, topology4.dyndns.atlas.ripe.net, topol-

gy6.dyndns.atlas.ripe.net, labs.ripe.net

dns, dns6 *.root-servers.net: TCP (SOA), UDP (SOA, version.bind, hostname.bind, id.server, version.server) sslcert, sslcert6 www.ripe.net, atlas.ripe.net http, http6 www.ripe.net/favicon.ico, ip-echo.ripe.net

▶ However, vantage point selection was based on simplistic fjlters that used −

1. Latitudes and longitudes
2. Network prefjxes

3 / 15

SLIDE 4

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

Introduction | Research Question

Jul ′14 Feature to add tags to probes [6]. Oct ′14 Feature to select probes using tags.

RIPE Atlas Probe Coverage (May 2017) https://atlas.ripe.net/results/maps/network-coverage

▶ How do we select dual-stacked vantage points at home? Do tags help? ▶ What region− and network−based bias comes into play with this selection?

4 / 15

SLIDE 5

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

System Tags

▶ Tags applied by RIPE Atlas based on results from built-in measurements.

3K 6K 9K 12K

# (Probes)

system-resolves-aaaa-incorrectly system-dns-problem-suspected system-resolves-a-incorrectly system-flakey-connection system-auto-geoip-country system-ipv4-doesnt-work system-doesnt-resolve-a system-doesnt-resolve-aaaa system-auto-geoip-city system-anchor system-resolver-mangles-case system-ipv6-ula system-ipv6-doesnt-work system-v1 system-v2 system-ipv6-works system-ipv6-capable system-v3 system-ipv4-rfc1918 system-resolves-aaaa-correctly system-resolves-a-correctly system-ipv4-works system-ipv4-capable 1 1 1 7 47 63 75 142 160 234 407 433 501 767 1427 3050 3640 6660 6715 8236 8305 8743 9045

▶ Frequently updated (every 4 hours) ▶ Fairly stable and accurate. ▶ Highlights state of DNS and IP connectivity. ▶ Helps identify hardware version of the probe.

5 / 15

SLIDE 6

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

System Tags | Load by Probe Hardware [7]

probev1 probev2 probev3

0.2 0.4 0.6 0.8 1 1 2 3 4 5 CDF Latency to first hop (ms) RESIDENTIAL PROBES Probev1 (144) Probev2 (313) Probev3 (963)

▶ v3 home probes show < 1 ms hop1 latencies. ▶ v1 and v2 probes show higher hop1 latencies (experience load issues).

6 / 15

SLIDE 7

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

Dual-Stacked Probes | Evolution

▶ ∼2.4K (∼25%) probes are dual-stacked (May 2017). ▶ Richest source of vantage points for IPv6 measurement studies. 2010 2011 2012 2013 2014 2015 2016 2017 2K 4K 6K 8K 10K API TAGS Connected, Non-Anchored Probes all dual-stacked ▶ Criteria:

1. Probes with same1 IPv4 and IPv6 ASN.
2. Probes with system-ipv4-works and system-ipv6-works tags.

1restrictive, but fjlters out hosts that use 6in4 tunnels 7 / 15

SLIDE 8

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

Dual-Stacked Probes | Country-based Distribution

▶ RIPE & ARIN regions with 91% probes. ▶ 88 countries / 822 ASNs covered. DE 489 US 304 FR 248 GB 161 NL 151 DTAG 181 COMCAST 169 PROXAD 96 XS4ALL 71 ORANGE 71 ▶ Belgium2 missing in the top 5 list! ▶ Probe deployment likely does not refmect

IPv6 user population across the globe.

Click the slices to view versions. Source: netmarketshare.com. IT: 1.3% IT: 1.3% IT: 1.3% SE: 1.3% SE: 1.3% SE: 1.3% GR: 1.4% GR: 1.4% GR: 1.4% FI: 1.4% FI: 1.4% FI: 1.4% AT: 1.4% AT: 1.4% AT: 1.4% NO: 1.8% NO: 1.8% NO: 1.8% CA: 1.9% CA: 1.9% CA: 1.9% RU: 2.2% RU: 2.2% RU: 2.2% CZ: 2.3% CZ: 2.3% CZ: 2.3% BE: 2.8% BE: 2.8% BE: 2.8% CH: 3.8% CH: 3.8% CH: 3.8% NL: 6.6% NL: 6.6% NL: 6.6% GB: 7.0% GB: 7.0% GB: 7.0% FR: 10.8% FR: 10.8% FR: 10.8% US: 13.2% US: 13.2% US: 13.2% OTHERS: 14.9% OTHERS: 14.9% OTHERS: 14.9% DE: 21.3% DE: 21.3% DE: 21.3%

Hig

Entire list at: http://goo.gl/UdEe1Q

2Belgium with ∼49% penetration leads (as of May 2017) Google IPv6 adoption statistics [8] 8 / 15

SLIDE 9

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

Dual-Stacked Probes | Underrepresented Countries

▶ Correlation of IPv6 users against dual-stacked probes using APNIC dataset [9] 20 40 60 80 Country Rank 0% 10% 20% 30% 40% 50% 60% 70% IPv6 Penetration Probes Users USERS PROBES BE 57.4% 2.8% LU 34.2% 0.6% GR 33.7% 1.4% CH 34.3% 3.8% PT 29.2% 0.7% IN 22.0% 0.1% US 33.2% 13.2% EC 18.8% 0.1% DE 39.9% 21.3% JP 19.8% 1.4% ▶ JP with ∼22M IPv6 users (∼19% IPv6 usage ratio) hosts only ∼1.4% (31) probes. ▶ Tiese countries with large IPv6 userbase can benefjt from more probes.

9 / 15

SLIDE 10

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

Dual-Stacked Probes | Network Type Distribution

▶ We further used PeeringDB [10] to map ASNs by network type.

2010 2011 2012 2013 2014 2015 2016 2017 101 102 103 104 # (Probes) TAGS Dual-Stacked Probes

ISP/NSP CONTENT PROVIDERS EDUCATIONAL/RESEARCH NON-PROFITS ENTERPRISE

NSP 1540 83.0% CP 139 7.5% EDU 110 6.0% NP 45 2.4% EP 21 1.1% ▶ ∼83% of dual-stacked probes are hosted by ISPs. ▶ ∼60% (782 probes) deployed at home. DSL 262 RESIDENTIAL CABLE 148 FIBRE 179

10 / 15

SLIDE 11

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

IPv4 versus IPv6 | Who connects faster?

−3 −2 −1 1 2 3 (slow-fast)/fast 1M 2M 3M 4M 5M 6M 7M # (Observations) IPv4 faster IPv6 faster Anchoring Measurements

▶ ping measurements from dual-stacked probes towards 149 RIPE Atlas anchors. ▶ 5th percentile latency comparison using month-long dataset (20M samples). ▶ Latency is comparable across AF, although IPv4 is marginally faster. ▶ Raw dataset is publicly released − http://goo.gl/dOJL5Q.

11 / 15

SLIDE 12

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

IPv4 versus IPv6 | Who connects faster?

▶ We used the identifjed dual-stacked probes to create a web service. ▶ For a given website (and origin-AS), it shows which AF connects faster.

http://goo.gl/hbzbwD

12 / 15

SLIDE 13

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

User Tags

▶ Tags manually applied by probe hosts. ▶ Requires proactive participation of probe hosts.

2010 2011 2012 2013 2014 2015 2016 2017 100 101 102 103 104 TAGS RIPE Atlas Probes

nat home ipv4 no-nat fibre cable ipv6 dsl native-ipv6

ffice

▶ Popular user tags are centered around home probes.

2K 4K

# (Probes)

ziggo fttc nat64 internet2 ftth dn42 wimax freifunk satellite 3g

range

fios cgn twc hackerspace mobile xs4all 4g known-ipv4-issues free ds-lite lte 6rd sixxs 6to4 upc vpn noc double-nat known-ipv6-issues ixp he wireless-isp dtag iwantbcp38compliancetesting comcast isp vdsl2 vdsl academic adsl multihomed core ipv6-tunnel native-ipv4 datacentre

ffice

native-ipv6 dsl ipv6 cable fibre no-nat ipv4 home nat 5 5 6 6 6 7 9 9 10 12 13 19 20 21 22 24 24 25 26 26 27 30 30 38 39 42 42 44 46 48 51 60 64 65 90 111 145 159 166 202 222 244 298 300 469 636 646 746 789 843 865 1061 1410 1583 2459 3934

13 / 15

SLIDE 14

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

User Tags | Tag Updates

User tags tend to become stale over time!

50 100 150 200 250 Frequency of Tag Updates 0.0 0.2 0.4 0.6 0.8 1.0 CDF sys user ['14-'17] Probes: 15.8K

▶ ∼2.8% of probe hosts ever update their user tags. ▶ ∼61.4% of probes received at least 1 update on system tags. ▶ ∼13.1% of probes received at least 10 updates on system tags.

14 / 15

SLIDE 15

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

Takeway

▶ User tags tend to become stale over time.

▶ Associate a tag-creation timestamp (allow predictive weighting of tag accuracy) ▶ Use built-in measurements to identify if user-tag is plausible.

▶ System tags (refresh every 4 hours) are stable and accurate. ▶ We used system tags to identify dual-stacked probes:

▶ 2.4K (∼25%) probes covering 88 countries and 822 ASNs. ▶ 83% deployed by ISPs with 782 connected at homes. ▶ Evenly split across DSL, cable and fjbre deployments. ▶ Some countries (such as BE and JP) are underrepresented in this sample.

www.vaibhavbajpai.com

bajpaiv@in.tum.de | @bajpaivaibhav

15 / 15

SLIDE 16

Introduction System Tags Dual-Stacked Probes

By region By network

Who connects faster? User Tags Takeway

References

[1] “RIPE Atlas: A Global Internet Measurement Network,” ser. Internet Protocol Journal (IPJ) ’15, September 2015, http: //ipj.dreamhosters.com/wp-content/uploads/2015/10/ipj18.3.pdf. [2]

V. Bajpai and J. Schönwälder, “A Survey on Internet Performance

Measurement Platforms and Related Standardization Efgorts,” ser. IEEE Communications Surveys and Tutorials (COMST) ’15, 2015. [Online]. Available: http://dx.doi.org/10.1109/COMST.2015.2418435 [3] “RIPE Atlas - Probe API: v1,” https://atlas.ripe.net/api/v1/probe, [Online; accessed 06-November-2015]. [4] “RIPE Atlas - Probe Archive API: v1,” https://atlas.ripe.net/api/v1/probe-archive, [Online; accessed 06-November-2015]. [5] “RIPE Atlas - Measurement Creation API,” https://atlas.ripe.net/docs/measurement-creation-api, [Online; accessed 06-November-2015]. [6] “RIPE Atlas - Midsummer Update 2014,” https://labs.ripe.net/ Members/fatemah_mafi/ripe-atlas-midsummer-update-2014, [Accessed: 04-Apr-2016]. [7]

V. Bajpai, S. J. Eravuchira, and J. Schönwälder, “Lessons Learned

From Using the RIPE Atlas Platform for Measurement Research,” ser. Computer Communication Review (CCR) ’15, 2015, pp. 35–42. [Online]. Available: http://doi.acm.org/10.1145/2805789.2805796 [8] “Google - IPv6 Adoption Statistics,” http://goo.gl/kKYXqS, [Online; accessed 22-Jan-2016]. [9] “ APNIC - IPv6 users by country,” http://labs.apnic.net/dists/v6dcc.html, [Online; accessed 22-Jan-2016]. [10]

A. Lodhi, N. Larson, A. Dhamdhere, C. Dovrolis, and kc clafgy,

“Using peeringDB to understand the peering ecosystem,” ser. Computer Communication Review (CCR) ’14, 2014, pp. 20–27. [Online]. Available: http://doi.acm.org/10.1145/2602204.2602208 15 / 15