1) The document summarizes measurements and lessons learned from World IPv6 Day. It discusses measurements of DNS records, ping/traceroute responses, and HTTP performance over IPv4 and IPv6 from vantage points to participating sites. 2) Key lessons included controlling DNS by setting low TTLs, testing deployments to avoid issues, and monitoring infrastructure to ensure IPv6 services work as expected and are withdrawn cleanly after events. 3) The measurements showed that negative caching and low TTLs were important for rapid propagation of DNS changes for World IPv6 Day, and that sites needed to monitor for issues with IPv6 connectivity even after events.

1. World IPv6 Day -
What did we learn?
emile.aben@ripe.net
APNIC 32

2. RIPE NCC Measurements - World IPv6 Day
• IPv6 Eyechart and 6to4 (not in this talk)
• Active measurements
– Sources: 40 vantage points (RIPE TTM, CAIDA Ark, ...)
– Destinations: 53 participant or already dual-stacked
sites
– From 2011-06-01 to 2011-06-11 we measured
– DNS: A and/or AAAA records
– ping(6)/traceroute(6)
– HTTP over IPv4 and IPv6
Emile Aben, APNIC 32 2

3. Measurement vantage points
Emile Aben, APNIC 32 3

4. Lesson:
Control

5. When did World IPv6 Day start?
• Less then 2 days before World IPv6 Day:
DNS negative caching for websites participating in World IPv6 Day
90
Number of participating websites
80
70
60
50
40
30
20
10
0
1m 2m 5m 10m 20m 1h 2h 5h 10h 1d 2d
minimum of: minimum TTL in SOA, TTL of SOA
Emile Aben, APNIC 32 5

6. Percentage of vantage points seeing AAAA
v6day
Emile Aben, APNIC 32 6

7. Percentage of vantage points seeing AAAA
v6day
Negative caching
Emile Aben, APNIC 32 6

8. Percentage of vantage points seeing AAAA
v6day
TTL caching
Negative caching
Emile Aben, APNIC 32 6

9. Control - lessons learned
• Know your on/off switch
– Control your DNS
• Set low TTL in case of
roll-back
• Set low negative TTL
– Bymeans of minTTL in
SOA / TTL of SOA
Emile Aben, APNIC 32 7

10. Lesson:
Test and monitor

11. You don’t want this to happen
On IPv4:
On IPv6:
Emile Aben, APNIC 32 9

12. Or this ...
Comparing DNS,ping and HTTP IPv6 measurements
to www.commerce.gov from 2011-06-09 0:00 UTC to 2011-06-10 12:00 UTC
100 %
% AAAA queries returning an IPv6 address
% HTTP responses == 200
80 % % ICMPv6 ping success
60 %
40 %
20 %
0%
00
04
08
12
16
20
00
04
08
12
:0
:0
:0
:0
:0
:0
:0
:0
:0
:0
0
0
0
0
0
0
0
0
0
0
Emile Aben, APNIC 32 10

13. Or this ...
Comparing DNS,ping and HTTP IPv6 measurements
to www.commerce.gov from 2011-06-09 0:00 UTC to 2011-06-10 12:00 UTC
100 %
% AAAA queries returning an IPv6 address
% HTTP responses == 200
80 % % ICMPv6 ping success
60 %
40 %
20 %
0%
00
04
08
12
16
20
00
04
08
12
:0
:0
:0
:0
:0
:0
:0
:0
:0
:0
0
0
0
0
0
0
0
0
0
0
v6 on www turned off
at end of v6day
Emile Aben, APNIC 32 10

14. Or this ...
Comparing DNS,ping and HTTP IPv6 measurements
to www.commerce.gov from 2011-06-09 0:00 UTC to 2011-06-10 12:00 UTC
100 %
% AAAA queries returning an IPv6 address
% HTTP responses == 200
80 % % ICMPv6 ping success
60 %
40 %
20 %
0%
00
04
08
12
16
20
00
04
08
12
:0
:0
:0
:0
:0
:0
:0
:0
:0
:0
0
0
0
0
0
0
0
0
0
0
v6 on www turned off but v6 service still
at end of v6day announced
Emile Aben, APNIC 32 10

15. Test and monitor - lessons learned
Emile Aben, APNIC 32 11

16. Test and monitor - lessons learned
• Test when deploying something
– The more real-life, the less likely you !@#$-ed up
• Monitor your infrastructure
• People2people reachability
– Avoidable situations like Level3 and Dept. Commerce
– Contact info up to date in RIR databases (whois)
– Monitor the web (NANOG, *NOG, Twitter, ...)
Emile Aben, APNIC 32 11

17. Global view

18. Performance of src/dst pairs on 2011-06-08
Distribution of IPv4/IPv6 relative performance
• Bell-shaped
IPv4 performs better IPv6 performs better
– centered around 0
– with fatter IPv4-side
• Dual-stack = two
Frequency
chances for best
performance!
– Real-time apps can
exploit this
(voice, gaming)
+ 400 %
+ 350 %
+ 300 %
+ 250 %
+ 200 %
+ 150 %
+ 100 %
+ 50 %
+ 40 %
+ 30 %
+ 20 %
+ 10 %
+0%
+ 10 %
+ 20 %
+ 30 %
+ 40 %
+ 50 %
+ 100 %
+ 150 %
+ 200 %
+ 250 %
+ 300 %
+ 350 %
+ 400 %
best performing protocol, as percentage−wise increase over the least performing protocol
Emile Aben, APNIC 32 13

19. Partial Reachability
• Internet is a collection of interconnecting
networks, and can be different on IPv4/IPv6
• For some of our vantage points, not all
destinations were reachable
– Are our vantage points representative?
– Network partitioning, examples we encountered:
– Level3 - Hurricane Electric
– Cogent - Hurricane Electric
– See http://en.wikipedia.org/wiki/Comparison_of_IPv6_support_by_major_transit_providers
Emile Aben, APNIC 32 14

20. Long term effects - Content
Percentage of web sites in Alexa 1M that can be reached over IPv6
4%
Alexa 1M
3.5 % Alexa top 25k
3%
2.5 %
2%
1.5 %
1%
0.5 %
0% 2 20 20 20 20 20 20 20 20
01 11 11 11 11 11 11 11 11
1- -0 -0 -0 -0 -0 -0 -0 -0
01 2 3 4 5 6 7 8 9
Raw data: Dan Wing
(http://banjo.employees.org/~dwing/aaaa-stats.html)
Emile Aben, APNIC 32 15

21. Long term effects - Content
Percentage of web sites in Alexa 1M that can be reached over IPv6
4% 8 June:
Alexa 1M
3.5 % Alexa top 25k 3.8%
3%
2.5 %
2%
1.5 %
1%
0.5 %
0% 2 20 20 20 20 20 20 20 20
01 11 11 11 11 11 11 11 11
1- -0 -0 -0 -0 -0 -0 -0 -0
01 2 3 4 5 6 7 8 9
Raw data: Dan Wing
(http://banjo.employees.org/~dwing/aaaa-stats.html)
Emile Aben, APNIC 32 15

22. Long term effects - Content
Percentage of web sites in Alexa 1M that can be reached over IPv6
4% 8 June:
Alexa 1M
3.5 % Alexa top 25k 3.8%
3%
2.5 % Pre/post 8 June:
0.4% -> 0.7%
2%
1.5 %
1%
0.5 %
0% 2 20 20 20 20 20 20 20 20
01 11 11 11 11 11 11 11 11
1- -0 -0 -0 -0 -0 -0 -0 -0
01 2 3 4 5 6 7 8 9
Raw data: Dan Wing
(http://banjo.employees.org/~dwing/aaaa-stats.html)
Emile Aben, APNIC 32 15

23. Long term effects - Content
Percentage of web sites in Alexa 1M that can be reached over IPv6
4% 8 June:
Alexa 1M
3.5 % Alexa top 25k 3.8%
3%
2.5 % Pre/post 8 June:
0.4% -> 0.7%
2%
1.5 %
1% Around 3 July:
0.7% -> 1.0%
0.5 %
(Single hoster in DE)
0% 2 20 20 20 20 20 20 20 20
01 11 11 11 11 11 11 11 11
1- -0 -0 -0 -0 -0 -0 -0 -0
01 2 3 4 5 6 7 8 9
Raw data: Dan Wing
(http://banjo.employees.org/~dwing/aaaa-stats.html)
Emile Aben, APNIC 32 15

24. Long term effects - Content
• Linear extrapolation:
–~ an IPv6 year needed to get to 100%
• Exponential extrapolation:
–~ an IPv6 week needed to get to 100%
Emile Aben, APNIC 32 16

25. Long term effects - Content
• Linear extrapolation:
–~ an IPv6 year needed to get to 100%
• Exponential extrapolation:
–~ an IPv6 week needed to get to 100%
• Note: extrapolation based on two data points is
not very scientific :)
Emile Aben, APNIC 32 16

26. Long term effects - IPv6 ASes
12 % 8 June
announcing IPv6 pfx
11.5 %
% of ASes
11 %
10.5 %
10 %
9.5 %
50
Daily increase
40
of IPv6 ASes
30
20
10
0
-10
-20
-30
20
20
20
20
20
20
11
11
11
11
11
11
-04
-05
-06
-07
-08
-09
-01
-01
-01
-01
-01
-01
• Higher growth before v6d: Deployments pushed earlier?
• Lower growth after v6d: Summer-vacation?
• http://v6asns.ripe.net
Emile Aben, APNIC 32 17

27. More information
• Web interface to the measurements
– http://v6day.ripe.net/
• Analysis on RIPE Labs
– http://labs.ripe.net/ipv6day
• Raw data availability
– http://labs.ripe.net/datarepository/data-sets/ripe-ncc-
active-measurements-of-world-ipv6-day-dataset
Emile Aben, APNIC 32 18

28. Conclusions - what we learned
• IPv6/dual-stack works just fine, but make sure
that
– It is properly tested and monitored (like IPv4)
– Your network can reach all others (like IPv4)
• Dual-stack = Two chances for best performance
• Days like this ‘work’
– Raise awareness
– Give people a target to work towards
– We’re ready for a next IPv6(day|week|month|year|∞)
Emile Aben, APNIC 32 19

29. Questions?
질문있습니까?