Re: initial issues

[itojun@iijlab.net]

>* There was a brief behind-the-scenes discussion, about the definition
>  of multihoming, which seemed to result in "being logically connected 
>  to more than one TLA or NLA simultaneously".

	i guess there still are many different configurations possible.
	we need to define what "multihome" means in this mailing list.

	the following tries to summarize what we discussed in the past,
	in couple of occasions (incl, ipngwg tokyo interim meeting and japan-
	local multihome workshop).

itojun


who are you?
- a SLA, or a leaf site (/48)
- an NLA, or small ISP (/n, where n < 48)
- a TLA, or big ISP (/16, /29-35 sTLA, or /24-28 pTLA)

goals of multihoming
- cope with L2 failures
- cope with upstream ISP failures
- load balancing, try to fill up two pipes we have toward upstream
	this needs more routing tricks.
- whatever you name it (but shouldn't dream too much, we need a workable
  operational solution not a 20-year-to-deploy new protocol)

backbone aggregation and routing table size
- RFC2373 and RFC2772 are pretty clear (for me) about aggregation is required
  when TLA/pTLA/sTLA site propagates their TLA/pTLA/sTLA to the backbone (DFZ).
  do we want to keep this property (pros: less specific routes  cons: no
  provider-independent, or TLA-independent address), or do we want to forget
  about it (pros: IPv4-like "i have provide-independent address and advertise
  it to both ISP" multihome possible  cons: more specific routes)
- what is the permissible maximum # of IPv6 routes in DFZ?  what is the
  limiting factor? (memory bound, CPU bound)  how can we enforce the max #?

for a SLA (/48 leaf site), multihome can be:
- connect to the same upstream ISP with multiple links
- connect to different ISPs with multiple links.  it has two axis (not very
  independent, and there's relationship with backbone aggregation):
  (1) ISP A and B belong to the same aggregated address block (like the same
      TLA).
  (2) ISP A and B belong to different aggregated address block (like the
      different TLAs).
  (a) SLA gets single /48 prefix from one of the upstreams (ISP A).  advertise
      it to both ISP A and B -> how far they will get propagated?
  (b) SLA gets two /48 prefixes from both of the upstreams (ISP A and B).
      advertise prefix A to ISP A only, advertise prefix B to ISP B only.

for an NLA?
for a SLA which is connected to multihomed NLA?
for a TLA?
- TBD

tools in our hand:
- DNS
- RFC2260-like configuration (tunnel, redundant L2 connectivity)
- multiple address assignment to a site
- source address selection logic at the end node
	note: unlike IPv4, every IPv6 node can handle multiple IP address
	assigned to a single interface.
- GSE-like mechanism?
- router renumbering protocol
- IPv6 autoconfiguration, including "deprecated" address handling

other issues:
- what happens to established TCP sesssions when ?
- how an endpoint measure which address is better and which is worse?
  (need global routing knowledge, or keep monitoring different paths)