[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: Minutes / Notes
| > Unfortunately, we've put the IP address in the
| pseudo-header checksum,
| > so when your locator changes, your connection fails. ..
| > Creating an identifier space is one way.
| Alternately, changing the
| > pseudo-header would be another, but we discarded
| that as being even
| > more unlikely.
| Hmmm. Why would adding a new namespace be any easier than
| changing the TCP
Because the politics are easier. No, I don't like that, but
it's very clear to me that pushing rocks up steep hills is
harder than pushing rocks downhill. I'm learning to live
with the political realities that I don't like and still
make forward progress.
| In either event, if you're multihomed, and you
| want to interoperate
| with hosts elsewhere, they have to be modified before you
| can talk to them.
They have to be modified before they can provide connection
survivability. You still have basic connectivity.
| Actually, this leads me to think. If interoperation with
| unmodified hosts is
| important, then all this discussion about alternative
| namespaces to identify
| endpoints is moot. At that point, there are precisely two
| i) Use multiple addresses, and if the link whose address
| you're using for the
| connction fails, the connection fails.
| i) Wse multiple addresses, and if the link whose address
| you're using for the
| connction fails, use MIPv6 mechanisms to divert it to a
| backup address.
| Note that either of these can be used, and you can use
| whichever of these
| fits your needs best. This does tend to fill Brian's
| suggestion that we have
| a range of solutions available...
The design team came up with an alternative that did not
require a tunnel endpoint buried inside of an ISP network,
used multiple PA addresses, and provided for connection
survivability. It has the advantage(?) that locators and
identifiers are both syntactically IPv6 addresses so that
unmodified hosts can continue to function without difficulty.
Yes, there are many different approaches to a specific solution,
but the need to create the namespace and manage it is clearly
overhead. We don't wish to multiply this by having a variety
of implemented solutions. While that might ease the standards
body politics, that would be effectively undeployable and thereby