Simple Multihoming Experiment

1
Simple Multihoming Experiment

• draft-huitema-multi6-experiment-00.txt
• Christian Huitema, Microsoft
• David Kessens, Nokia

2
Simple dual homing problem statement

• Simple bridged network
• 2 routers, 2 ISP
• ingress filtering
• No ISP coordination
• Example
• T1 DSL back-up
• 2 DSL modems
• DSL cable
• Cable WiFi mesh
• Several hosts
• Simple IPv6 basic
• Advanced multi-homing aware
• It must work!

Internet (IPv6)
ISP1
ISP2
R1
R2
Single link (bridge)
H
H
H
3
IPv4 equivalent back-up

• Use for Back-Up
• Switch R1 on if R2 is down
• May incur small delay
• In general, loose TCP connections
• Typically combined with NAT DHCP
• Private addresses, no renumbering

Internet (IPv4)
ISP1
ISP2
R1
R2
Single link (bridge)
H
H
H
4
Broad Lines of the Solution

• No coordination between ISP
• Use of Provider Addresses
• Each ISP allocates a prefix
• or each ISP allocates an IPv4 address, and the
  routers use 6to4
• Multi-Addressing
• Each router advertises a prefix
• Hosts configure addresses with each prefix
• Five issues need resolution

5
Multi-addressing issues

• Ingress filtering
• host pick address from ISP1, send through R2?
• Dead default exit router, or dead ISP
• host keeps sending packets through a black hole?
• Inbound connection through wrong ISP
• Peers send packet to the black-holes address?
• Maintaining TCP connections
• Keep TCP going if the Router or the ISP fails?
• Use the right exit/entrance
• Maybe some amount of load balancing

6
Ideas, Ingress Filtering

• Choice by host
• Host treats multiple auto-config prefixes as
  sub-interfaces, associates individual IPv6
  address and default router.
• Easy to implement in new hosts.
• Redirect at routers
• No need for tunnels in single link network.
• Guarantees that old hosts keep working.
• There may be other solutions
• New services, ISP involvement, etc.

7
Ideas, dead exit router or dead ISP

• If the router notices the problem
• Advertises prefix as deprecated, or stop
  advertising
• Will not be used for new connections
• Will work for old and new hosts.
• If the problem is not really detected
• New host tries multiple source addresses when
  establishing a new connection
• Host may keep track of the quality of each router
  connection

8
Ideas, Peer using dead address

• If the problem is detected
• Update the name server?
• If the problem is not detected
• DNS advertises multiple addresses
• Peer tries several addresses
• Issue TCP timers?

9
Ideas, Maintaining TCP Connections

• No good solution for old hosts
• But there is no solution in a similar IPv4 set-up
  either
• Many applications will automatically reconnect
• New hosts may use MIPv6
• See Application of the MIPv6 protocol to the
  multi-homing problem
• draft-bagnulo-multi6-mnm-00
• SCTP may also be used
• See multi-homing issues in SCTP
• draft-coene-sctp-multihome-04.txt

10
Idea, Selecting the right exit/entrance

• Right entrance DNS tricks
• In asymmetric scenarios (back-up), only publish
  the best address in the DNS
• In symmetric scenarios, publish both
• Right exit Routing tricks
• Provide information in router announcement, as in
  Default Router Preferences, More-Specific
  Routes, and Load Sharing
• draft-ietf-ipv6-router-selection-02.txt

11
Summary

• It looks good on paper
• All issues have plausible solutions
• No change required to IPv6 standards
• No need to rewrite the IPv6 address at site exit
• We would benefit from mobile IPv6 and router
  selection work
• But we would like an actual deployment
• In theory, there is no difference between theory
  and practice, but in practice there is!

12
Range of solutions

• Small sites
• Do as we just explained
• Medium sites
• Ask the ISP to cooperate, allow both source
  addresses in ingress filtering
• E.g. add a local route to the other prefix
• Very large sites
• Treat as ISP, get their own prefix AS