Dale, I'm not sure your proposal is compatible with how the type/length field and ethertype field are distinguished. If the first octet is less than 0x06, then the field is a type/length field per 802.3.
https://standards.ieee.org/develop/regauth/tut/ethertype.pdf Paul On Sat, Mar 3, 2018 at 8:08 PM, Dale R. Worley <wor...@ariadne.com> wrote: > > Section 5.3 ("TCP/UDP over Geneve") describes an optimization that > > avoids sending the inner IP header. That optimization is not specific > > to packet spraying. The optimization needs to be negotiated between > > the Geneve endpoints or configured using an SDN controller, which is > > not mentioned in Section 5.3. Section 5.3 should be a separate I-D. > > draft-xiang-nvo3-geneve-packet-spray-00 handles putting a Layer 4 > protocol (TCP or UDP) on top of Geneve by assigning new Ethertypes > (0x9004/9005). A more elegant way is what I proposed last July: Due to > the different types of Ethernet, Ethertypes are restricted to 0x0600 and > higher; the values 00 through 05 can't appear as the first byte. > (Though there actually are IEEE assignments in that range, there are > none with a high byte of 00.) We can therefore modify the definition of > the "Protocol Type" field in the Geneve header such that if the first > byte is 00, then the second byte is the IP "protocol" value for the > contained packet. That covers TCP, UDP, and all Layer 4 protocols > carried inside of IPv4/IPv6. > > Dale >
_______________________________________________ nvo3 mailing list nvo3@ietf.org https://www.ietf.org/mailman/listinfo/nvo3
