On Fri, Oct 14, 2005 at 12:33:51PM -0700, william(at)elan.net wrote:
On Fri, 14 Oct 2005 bmanning@vacation.karoshi.com wrote:
Since shim6 requires changes in protocol stacks on nodes, my impression has been that it isn't a _site_ multihoming solution, but rather a _node_ multihoming solution. Is my impression incorrect?
There is no shortage of rough corners to file down, and I am behind on my shim6 mail, but the general idea is to let end sites multi-home in the "bag-o-PA-prefixes" style and let the nodes within that site use their multiple globally-unique addresses (one per upstream, say) to allow sessions to survive rehoming events.
the kicker here is that the applications then need some serious smarts to do proper source address selection.
No. The kicker is that the applications needs no such smarts and shim6 will take care of this for all applications on the system on the network level.
call me skeptical... I last followed this line of thinking w/ Rich Draves hammered out the issues of source address selection in his I-D's (and they might have become RFCs) and they are not trival. vendors getting this right, let alone interoperable will be a significant feat. Handwaving and Tapdancing not withstanding. Then there is the minor detail in a service provider figuring out what, if anything, she needs to do to migrate from the existing, timetested, working methods of multi homing to this new'nimproved method... reminds me of the old NewYorker cartoon; "...and then a miracle occurs..." --bill
-- William Leibzon Elan Networks william@elan.net