On 7/21/2016 1:22 PM, Dino Farinacci wrote: >> Dino, >> >> I'm curious - what is your basis for "cost”? > (1) Time to market due to complexity in chip design. > (2) More are used for logic versus needed SRAM space for table entries. > (3) Additional latency in processing packets. > >> The overheads in these protocols is as low as any other currently in >> widespread use. > Not at the data-plane. No one processes IPv4 options except for the RA > option. And that is why IPv4 goes fast. And IPv6 is following this path as > well but there are new proposals out that will slow it down due to more > processing. When I say “slow” here I mean the cost-effectiveness as described > in the 3 bullets above.
These are data plane functions for ingress/egress processing, which already does things like IPsec tunnel mode, IP in IP, IP in UDP in IP, etc. I'm not sure about the others, but why is GUE so complex? It would still allow data plane processing using existing IP forwarding, including DPI access to port numbers. Joe _______________________________________________ nvo3 mailing list [email protected] https://www.ietf.org/mailman/listinfo/nvo3
