[Bier] DISCUSS: suggestions/Q for draft-ietf-bier-mpls-encapsulation-02

[Toerless Eckert <[e-mail]>]

Couple of comments/Q:

Q1. draft-xu-bier-encapsulation vs. draft-ietf-bier-mpls-encapsulation

    Was this meant to be for BIER over non-MPLS, is this still pursued ?

C2: draft-ietf-bier-mpls-encapsulation-02

    IMHO, this can perfectly be used to carry BIER over non-MPLS,
    the first four bits even make it compatible with encaps without
    next-proto field that only support IPv4/IPv6.

    All the MPLS header elements are IMHO required for BIER-over-whatever.
    I don't think this header should be reinvented over and over again
    in other drafts. 
    
    If BIER-over-something-else requires ADDITIONAL header elements,
    eg: MT-ID or the like (as suggested by draft-xu), then these could
    go into extension headers. Simple way for extension header:
    Set "Proto" to "extension header N", define that extension header in
    another draft.

C3: The discussion with Caitlin re. BIER in DC brought up IMHO the
    cool thought of running BIER end-to-end into apps.

    Today, the MPLS header only support next-proto for "transport services",
    aka: encap of MPLS, IPv6/IPv6, Ethernet and for OAM. With these
    options, one would need to run eg: UDP/IP/BIER. IMHO, the IP
    header, especially IPv6 is fully redundant wastes 60? bytes and
    creates unnecessary confusion (what addresses to use). The sender/receiver
    IP addresses are simply the BFR-prefixes that are known from routing
    underlay without need to be carried in a packet header (BFIR-ID
    in packet indicates BFIR-prefix).

    Aka:
    a) Allocate one code-point for UDP.
    or:
    b) allocate a code-point for "IP-transport", and put the 8-bit
       ip transport protocol into the reserved field.
    
    If course, this is just a "would be cool to have". If somebody really
    wants to implement end-to-end BIER sockets, he could certainly manage to
    send/receive also a redundant IP header.

Cheers
    Toerless