Re: Multiple Address Service for Transport (MAST)

[Masataka Ohta <mohta@necom830.hpcl.titech.ac.jp>]

Dave;

MAST is another form of (original) LIN6.

While you say:

	It requires no change to the IP infrastructure, and no change
	to IP modules or transport modules in the end-systems.

	It requires no change to existing transport protocols and
	no changes to IP. Instead, it defines a wedge layer between
	IP and transport.

	In contrast, the current proposal provides support that
	requires no new infrastructure and no changes to existing
	protocols.

	The service should require no changes to the IP network
	infrastructure, to the IP layer in end-systems, or to the
	transport layer in the end-systems.

They are all wrong.

First, a transport protocol is defined on the wire, not at the
API. So, if, on the wire, packets with different pairs of
source/destination addresses belong to the same connection,
it is not TCP but a new protocol.

Second, while you don't have to modify existing transport
code, you must add something to the transport module. That is,
as transport has its own checksum on addresses, calculation
of which is transport dependent, your code include additional
transport module. If you want to add new transport, your
code must also be modified.

Yes, as you correctly observe, your proposal is only as
transparent as NAT.

You modify everything only to have a system to be able to
handle address changes not so efficiently.

But, the worst problem of your proposal is that you don't
properly address the most serious issue on when to change
addresses and which address should be preferred. Read my
draft.

							Masataka Ohta