IPv6 Deployment Case on a Korean Governmental Website, by Jean Ryu.
A presentation given at APNIC 42's Network Operations session on Tuesday, 4 October 2016.
20240509 QFM015 Engineering Leadership Reading List April 2024.pdf
IPv6 Deployment Case on a Korean Governmental Website
1.
2. National Internet Registry of Korea
• KrNIC
- ccTLD service and DNS (.kr, .한국)
- IP allocations v4&v6
- IDRC
- IPv6 Promotion & Support
• KrCert/CC
• and many more
http://kisa.or.kr
3. • Current IPv6 commercial services in Korea
- LTE mobile service (Sep 2014~), more than 6 million IPv6 devices
- CATV service (Dec 2015~), 60,000 IPv6 subscribers
- B2B, B2C internet, …?
- KISA is providing IPv6 internet connection for free of charge
• But nearly no contents…
6. • Trial IPv6 (dual stack) deployment on a mobile website of Korean
ministry – MSIP – for a month
• Korean governmental websites are being operated in the integrated
data center called NCIS
• Very sensitive and careful regarding service stability
‐ We decided not to change IPv4 systems at all and configure a new IPv6
based test network for the trial service
7. • Fixed line IPv6 internet service is not being provided yet…
Governmental
Data Center
IPv6
Internet
IPv4 ISP
IPv4 ISP
IPv4 ISP
6to4 Tunnel
9. • Resource synchronization (DB, etc)
• Some components without IPv6 support…
‐ Security equipments
‐ And softwares
10. • Some companies have announced that IPv6 performs better than
IPv4 in specific environments
• Could we expect reliable and fast IPv6 connection as well?
17. • The trial service was reliable enough
- no connection failure or user complaint
• However, IPv6 connection was notably slower than IPv4
- We found that in many cases, web browsers firstly tried to access the
website by IPv6, and then fallback to IPv4 because they weren’t able to
make the connection until the Happy Eyeball time limit…
• What makes this difference? Why IPv6 is slower and unreliable?
18. • Comparison of the number of BGP paths of IPv4 and IPv6
nearly no path!
(APNIC vizAS)
20. • We can hardly find IPv6 native paths in Korea
- A vast majority of IPv6 paths are 6in4/6to4 tunnel connected to KISA
- Moreover the difference in the number of BGP paths is enormous
- It’s difficult to fairly compare IPv6 and IPv4 in this situation, but it’s our
circumstance anyway
• Which factors does influence on the performance?
- The number of BGP paths? Transition technologies? NAT middle boxes?
Protocol itself? What else?
- Which are the factors we can control and which are the ones we can’t?
- That is to say, what can we do in order to improve this situation?
21. • We’re trying to arrange and persuade the top-tier ISPs’ IPv6 BGP
peering to make more IPv6 paths
• Currently the connection failure of IPv6 may possibly happen because
of too few paths
‐ Web browsers have Happy Eyeball fortunately, but there’re another
applications without fallback algorithms (which means higher risk)
• How shall we help the contents providers avoid these kind of
business risks and get rid of their anxieties of deploying IPv6?
‐ Fast fallback algorithm for all environments?
22. IPv6 Only
IPv4/IPv6
Dual Stack
IPv6 Only
eNodeB SGW PGW
L3 SW PE
NAT64
PE PE
Core RT
L2 SW L3 SW
NAT64
PECore RT
DNS
DNS64
DNS
DNS64
Core RT
IPv4
IPv6
u Packet Generator
u Frame Size Control
u Throughput Control
u IPv4/IPv6 Packet mix control
u traceroute
u Hop Count Check
u Hop by Hop Response Time
Check
u ping
u Response Time Check
u Web Browser
u Happy Eyeball check
u Round Trip Time check
u Wireshark
u Packet analysis
u Throughput Check
u Delay Check
u Frame Loss Check
u Round Trip Time Check
u Hop Count & route pachCheck
u Address Translation Performance Check
u DNS Response Time Check
u Top 100 Web Service Response
u Web Image Download Time
u File Download Time
u Video/Audio Streaming
u Game Contents Download
u VoIP