On IPv4 and IPv6 Rou.ng Stability Ioana Livadariu, Ahmed Elmokashfi, Amogh Dhamdhere Mo.va.on RIR#Address#Pool(/8s)###
•  IPv4 addresses are running out: -­‐  Feb 2011: IANA allocated last /8 -­‐  3 RIRs (APNIC,RIPE,LACNIC) are allocaKng from the last /8 Figure#no.1#:#IPv4#Address#Report#*#
•  Problem: Slow migraKon to IPv6 •  Look at performance by analyzing rou5ng stability Figure no.2 : IPv6 client-­‐side adopKon** * RIR IPv4 Address Run-­‐Down Model [ hRp://www.potaroo.net/tools/ipv4/] ** IPv6 Sta.s.cs [ hRp://www.google.com/intl/en/ipv6/staKsKcs.html] Measuring Rou.ng Stability : metrics Control plane ( BGP updates at 5 dual-­‐stacked RV monitors ) –  Frequency of rou5ng changes towards IPv4 and IPv6 prefixes –  Correla5on of the IPv4 and IPv6 rou5ng events Data plane ( 9 ARK monitors to probe dual-­‐stacked targets ) –  Reachability of the dual-­‐stacked probed targets –  RTT varia5ons IPv6 rou.ng system exhibits more rou.ng changes than IPv4 1
1
1
0.1
0.1
1
0.01
CCDF
CCDF CCDF
0.1
0.01
0.01
0.001
0.1
0.001
CCDF
0.01Jul10-IPV4
Jul12-IPV4
Jul10-IPV4Jul10-IPV4
0.0001
0.001
Jul12-IPV4Jul10-IPV6
0.0001
Jul10-IPV4Jul12-IPV6
0.001Jul14-IPV6
Jul10-IPV6
Jul10-IPV4
Jul12-IPV6
Jul12-IPV4
1e-05
Jul14-IPV6
Jul10-IPV4
1
10
100
1000
100
10000.0001
Jul10-IPV4
Jul10-IPV6
1e-05
Jul12-IPV4
daily convergence sequences per active prefix
ence sequences per active prefix
Jul12-IPV6
0.0001 # of
1
10 Jul10-IPV4
100
1000
100
1000
Jul14-IPV6
Jul10-IPV6
# of daily convergence
sequences per active prefix
uences per active prefix
1e-05
Jul12-IPV6
1
10 Jul14-IPV6
100
1000
100
1000
1e-05
# of daily convergence sequences per active prefix
uences per active prefix
Number'of'rou*ng'events'per'prefix'per'day'('VP:'NTT')''
1
10
100
10
100
1000
# of daily convergence sequences per active prefix
onvergence sequences per active prefix
1000
•  IdenKfied IPv4 and IPv6 rouKng events •  0.1% of the IPv4 versus 2% of the IPv6 prefixes experience more than 100 events per day Zooming in : Ac.ve prefixes •  Ac5ve prefix : a prefix that experiences a rouKng change at least once per day VP : HE VP : IIJ 0.12
0.09
0.11
0.08
0.07
0.09
Fraction of prefixes
Fraction of prefixes
0.1
0.08
0.07
0.06
0.05
0.06
0.05
0.04
0.04
0.03
0.03
IPv6
IPv4
IPv6
IPv4
0.02
0.02
09
10
11
12
Date
13
14
09
10
11
12
Date
13
14
Average fracKon of daily acKve prefixes •  The overall fracKon of acKve prefixes is higher in IPv6 than IPv4 Zooming in (more) : Highly ac.ve prefixes •  Highly ac5ve prefixes : top 1% of the acKve prefixes in terms of contribuKon to the BGP dynamics VP : NTT VP : HE 0.7
0.55
0.5
Contribution to overall activity
Contribution to overall activity
0.6
0.5
0.4
0.3
0.45
0.4
0.35
0.3
0.25
0.2
0.2
IPv6
IPv4
0.1
IPv6
IPv4
0.15
09
10
11
12
Date
13
14
09
10
11
12
Date
13
14
The contribuKon of the top 1% acKve prefixes to the BGP dynamics •  The highly acKve prefixes are responsible for between 40-­‐50% of IPv6 updates compared to 20-­‐30% of IPv4 updates Correla.ng IPv4 and IPv6 rou.ng events VP: APAN 0.6
0.55
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
Avg congruence per AS
Avg congruence per AS
VP : HE 09
10
11
12
Date
13
Identical v4/v6 paths
Different v4/v6 paths
14
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
09
10
11
12
Date
13
14
Identical v4/v6 paths
Different v4/v6 paths
•  Computed the fracKon of overlapping IPv6 and IPv4 events •  Higher correlaKon for idenKcal paths than for different paths Approach to study Data plane Stability •  Goal : Study reachability and performance •  Measurement setup: Use nine monitors from the ARK infrastructure to ping dual-­‐stacked targets* every 5 seconds ( ~105 targets per monitor ) for 1 ½ months ( August – September 2014); •  Limita5ons : Data could be influenced by the availability of the webservers and random losses; •  Data Filtering : Filter out very long or very short periods of unreachability( [15s , 3600s ]) Alexa [hRp://www.alexa.com/topsites ] *_ Network reachability over IPv4 than IPv6 1
0.9
0.8
CDF
0.7
0.6
0.5
0.4
0.3
0.2
0.1
IPv4
IPv6
0
101
102
103
104
Length of unreachability time window (seconds)
•  Longer unreachability intervals over IPv6 than IPv4 •  Paired measurements : The difference in unreachability towards the same AS can be up to 15% Performance : Similar RTT .me series IPv4 IPv6 Monitor : ams-­‐nl; Target : AS 197043 Performance : Discrepancies in the RTT .me series IPv6 IPv4 Monitor : ams-­‐nl; Target : AS 15982 Conclusions & Future work Conclusions ( so far ) •  Control plane: IPv6 prefixes are less stable than IPv4 Most IPv6 rouKng dynamics are generated by a small fracKon of pathologically unstable prefixes •  Data plane: IPv6 unreachability intervals longer and more frequently than IPv4 unreachability intervals Ongoing work •  Use the same measurement setup to collect traces of the paths between the ARK monitors and the probed webservers •  Correlate the control and data plane events