Required attributes for transport layer protocols on Internetworks
Andre Oppermann
nanog-list at nrg4u.com
Fri Oct 14 16:36:07 UTC 2005
Independent of all this discussion we are witnessing regarding to the IPv6
deployment I'd like to write down some high-level requirements for transport
layer protocols in Internetworks (such as the global Internet).
Lets have a look at required attributes of such an ideal transport layer
protocol for Internetworks without regard to any existing protocols:
1) It should have sufficient address space to allow every human and corporation
on this planet to have all his/her electronic devices online with a unique
address at the same time.
[Large address space.]
2) It should have indivdually routeable fixed- or variable sized netblocks.
[No individual number allocation. Aggregation of individual addresses
into routable blocks.]
3) It should function in a global dynamic and automatic routing system based
on netblocks.
[BGP or better.]
4) It should have routing of any netblock independent of political or national
hierarchies or assignments.
[Aggregation follows the network structure, not political/geographical borders.]
5) It should have globally unique netblocks which are allocated to connectivity
providers who then redistribute/assign parts to end user connectivity.
[The delta between address/netblock per individual user vs. routable netblock
should be larger than 2^12. Provider aggregation.]
6) It should have globally unique netblocks which are independent of connectivity
providers and assigned directly to end users with sufficient requirements.
[Provider independent address space. This address space may or may not be
directly routable.]
7) It should have minimum requirements for netblocks in size and aggregation to
participate in the global routing system.
[Minimum allocation.]
8) It should restrict itself entirely to the transport layer (OSI layer 3).
[No source routing, flow labels and such. That's the job of layer 2.5.]
9) It should only be used for transport layer purposes.
[Only for packet forwarding. No additional meaning as in phone numbers, etc.]
10) KISS. Keep it simple, stupid.
[If it takes $10k worth of courses to understand it's unsuitable.]
When going through this list we see a couple of points in which IPv4 and IPv6
fail miserably.
Food for thought. Get the discussion started.
--
Andre
More information about the NANOG
mailing list