Measuring IPv6 Adop3on
Jakub Czyz, University of Michigan Mark Allman, Interna=onal Computer Science Ins=tute
Jing Zhang, University of Michigan ScoA Iekel-‐Johnson, Arbor Networks
Eric Osterweil, Verisign Labs Michael Bailey, University of Michigan and University of Illinois
SIGCOMM 2014
Chicago, IL, USA August 17-‐22, 2014
Why Study IPv6 Adop3on Now?
SIGCOMM’14 Measuring IPv6 Adop=on – Czyz et al. 2
(Image source: Geoff Huston, hAp://www.potaroo.net/tools/ipv4)
• Internet con=nues growing • IPv4 space shrinking… • IPv4 exhaus=on events:
– IANA: February 2011 – Asia/Pacific: April 2011 – Europe: September 2012 – La=n America: June 2014
• IPv6 Community Flag Days – 2011 & 2012
Total Free IPv4 /8 At Registries
IANA Exhaus=on
Our Study
• Goal: a systemic “big picture” of IPv6 adop=on – Trading off depth for breadth – Are there cross-‐perspec=ve insights?
• Mul3-‐perspec3ve: 10 datasets • Mul3-‐year: 2-‐10 years • Mul3-‐aspect: 12 metrics • Findings: IPv6 adop3on – varies by where you measure (region) – varies by what you measure – recently made a qualita=ve jump
SIGCOMM’14 Measuring IPv6 Adop=on – Czyz et al. 3
Data Analyzed
• Exis3ng/Public Datasets: – RIR alloca=on – Route Views BGP, RIPE-‐RIS BGP – Google.com clients, – Verisign zone files, – CAIDA Ark RTT
• New Datasets: – Traffic: Arbor Networks global traffic – Naming: Verisign .com/.net queries via IPv4, via IPv6 – Content: Tes=ng data of Alexa top-‐10K sites
SIGCOMM’14 Measuring IPv6 Adop=on – Czyz et al. 4
Metrics Prerequisite IP Func3ons • Address Alloca=on • Address Adver=sement • Topology • DNS Name servers • DNS Resolvers • DNS Queries • Server Readiness • Client Readiness
SIGCOMM’14 Measuring IPv6 Adop=on – Czyz et al. 5
Opera3onal Characteris3cs • Traffic Volume • Applica=on Mix • Transi=on Technologies • Performance (RTT)
“IPv6 adop
3on” = lev
el rela3ve
to IPv4
METRICS (PREREQUISITE)
SIGCOMM’14 Measuring IPv6 Adop=on – Czyz et al. 6
Prefix Alloca3on
0
200
400
600
800
1000
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 0
0.1
0.2
0.3
0.4
0.5
0.6
Mon
thly
Pre
fix A
lloca
tions
Rat
io IP
v6/IP
v4
IPv4IPv6
Ratio
Measuring IPv6 Adop=on – Czyz et al. 7 SIGCOMM’14 IANA
Exhaus=on
Naming: Domains & Record Types
0
0.2
0.4
0.6
0.8
1
v4 v6 v4 v6 v4 v6 v4 v6 v4 v6
Frac
tion
of A
ll D
NS
Que
ries
2011ï06ï08 2012ï02ï23 2012ï08ï28 2013ï02ï26 2013ï12ï23
other ANYTXTNSDSMXAAAA A
Measuring IPv6 Adop=on – Czyz et al. 8 SIGCOMM’14
More Similar (p < 0.05)
• Queries from .com/.net; IPv4 & IPv6 name servers – Five day-‐long packet samples over 2.5 years – IPv6 DNS users query similar domains as IPv4 – Query types are converging over this =me period:
Server Readiness: Alexa Top Domain Reachability
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
2011-06 2011-12 2012-06 2012-12 2013-06 2013-12
Frac
tion
of A
lexa
Top
10K
AAAA LookupsReachability
Measuring IPv6 Adop=on – Czyz et al. 9 SIGCOMM’14
IPv6 World Day IPv6 Launch Day
Client Readiness: visitors to google.com
0
0.005
0.01
0.015
0.02
0.025
2009 2010 2011 2012 2013 2014
Frac
tion
Clie
nts
Usi
ng IP
v6
Measuring IPv6 Adop=on – Czyz et al. 10 SIGCOMM’14
(Data method in Colir et al., 2010)
+151%
+147%
+61% -‐7% +43%
METRICS (OPERATIONAL)
SIGCOMM’14 Measuring IPv6 Adop=on – Czyz et al. 11
Global Traffic
10M
100M
1G
10G
100G
1T
2010 2011 2012 2013 2014 0.0001
0.001
0.01
Traf
fic V
olum
e/C
usto
mer
(bps
)
Rat
io IP
v6/IP
v4
0.0064
IPv4 A (peak)IPv6 A (peak)
Ratio A (peaks)
IPv4 B (average)IPv6 B (average)
Ratio B (averages)
Measuring IPv6 Adop=on – Czyz et al. 12 SIGCOMM’14
• Arbor Networks global provider neslow data • 260 service providers (Dataset B) ~ 1/3 – 1/2 of all inter-‐AS traffic
+433% +470%
Applica3on Mix (% of IPv6)
Measuring IPv6 Adop=on – Czyz et al. 13 SIGCOMM’14
{ User content
IPv6 Transi3on Technologies (Teredo + 6to4)
0
0.2
0.4
0.6
0.8
1
2009 2010 2011 2012 2013 2014
Frac
tion
of n
on-n
ativ
e IP
v6
Internet Traffic AInternet Traffic B
Google Clients
Measuring IPv6 Adop=on – Czyz et al. 14 SIGCOMM’14
Mostly Transi3on
Mostly Na3ve
CONCLUSIONS
SIGCOMM’14 Measuring IPv6 Adop=on – Czyz et al. 15
Conclusion 1: Regions Differ
1e-05
0.0001
0.001
0.01
0.1
1
Address(A1) Routing(T1) Traffic(U1)
AFRINICAPNIC
ARINLACNIC
RIPENCC
Measuring IPv6 Adop=on – Czyz et al. 16 SIGCOMM’14
IPv6 / IPv4 Ra=
o
} Large Inter-‐Region Differences
Large Intra-‐Region (Cross-‐Metric) Differences
E.g. ARIN last place in alloca3on, first in traffic.
Conclusion 2: Perspec3ve Magers
Measuring IPv6 Adop=on – Czyz et al. 17 SIGCOMM’14
0.0001
0.001
0.01
0.1
1
2009 2010 2011 2012 2013 2014 0.0001
0.001
0.01
0.1
1
IPv6
/ IP
v4 R
atio
A1 (allocation - monthly)
A1 (allocation - cumulative)
A2 (advertisement)
R2 (Google clients)
U1 (traffic - A.peaks)
U1 (traffic - B.averages)
N1 (.com NS)
T1 (topology)
P1 (performance)
2-‐3 order of magnitude difference
Conclusion 3: IPv6 is Real!
Measuring IPv6 Adop=on – Czyz et al. 18 SIGCOMM’14
ß 20x growth!
ß 15x growth!
ß Traffic Flipped
ß Nearly on-‐par
Thank You!
Ques=ons?
SIGCOMM’14 Measuring IPv6 Adop=on – Czyz et al. 19
BACKUP SLIDES
Measuring IPv6 Adop=on – Czyz et al. 20 SIGCOMM’14
Internet Devices and Users Con3nue to Increase
SIGCOMM’14 Measuring IPv6 Adop=on – Czyz et al. 21
Dataset Summary
Measuring IPv6 Adop=on – Czyz et al. 22 SIGCOMM’14
/day /day
Prefix Adver3sement
100
1K
10K
100K
1M
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 20140.0001
0.001
0.01
0.1
Pref
ix A
dver
tisem
ents
Rat
io IP
v6/IP
v4
IPv4IPv6
Ratio
Measuring IPv6 Adop=on – Czyz et al. 23 SIGCOMM’14
AS Centrality
0
2
4
6
8
10
12
14
16
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
K-co
re d
egre
e
Dual-StackIPv6-Only
IPv4
Measuring IPv6 Adop=on – Czyz et al. 24 SIGCOMM’14
DNS: .com & .net Zones
10
100
1K
10K
100K
1M
2007 2008 2009 2010 2011 2012 2013 2014 1e-05
0.0001
0.001
0.01
Rec
ords
in T
LD z
ones
Rat
io IP
v6/IP
v4 (.
com
)
.com A glue.net A glue
.net AAAA glue.com AAAA glueRatio .com glue
Ratio .net all probed (H.E.)Ratio .com all probed (H.E.)
Measuring IPv6 Adop=on – Czyz et al. 25 SIGCOMM’14
IPv4 .com/.net TLD A and AAAA Query Rank Correla3on (Spearman’s ρ)
Measuring IPv6 Adop=on – Czyz et al. 26 SIGCOMM’14
Within type: Strong Across type: Weak
Naming: Domains • Queries from .com/.net; IPv4 & IPv6 name servers – Five day-‐long samples over 2.5 years
• Four sets of top 100k domains: – For both IPv4 and IPv6 packets (user popula=ons) – Within each, for domains queried by A and AAAA
• Finding: IPv4 to IPv6 popula=ons correlate strongly for the same query type {A,AAAA} – e.g. Spearman’s ρ of 0.7 for IPv4 A versus IPv6 A
• So, IPv6 DNS users query similar domains as IPv4
SIGCOMM’14 Measuring IPv6 Adop=on – Czyz et al. 27
Performance (using 10-‐ and 20-‐hop RTT)
0
100
200
300
400
500
2009 2010 2011 2012 2013 2014 0.5
0.6
0.7
0.8
0.9
1
Med
ian
RTT
(ms)
Rat
io IP
v6/IP
v4 p
erfo
rman
ce
IPv6 - Hop 20IPv4 - Hop 20IPv6 - Hop 10IPv4 - Hop 10
Ratio - Hop 10
Measuring IPv6 Adop=on – Czyz et al. 28 SIGCOMM’14
Projec3ons
0.0001
0.001
0.01
0.1
1
2011 2012 2013 2014 2015 2016 2017 2018 2019
IPv6
/IPv4
A1 (allocation - cumulative)
U1 (traffic - A.peaks)
Polynomial Proj. (A1:R2 = 0.996; U1:R2 = 0.838)Exponential Proj. (A1:R2 = 0.984; U1:R2 = 0.892)
Measuring IPv6 Adop=on – Czyz et al. 29 SIGCOMM’14