The MPLS WG Archive

[Bora Akyol <akyol@pluris.com>]

Let me clarify myself, ifindex was "agreed" (AFAIR) to be 32 bits long to
better accommodate SNMP. Our internal implementation uses 32
bits for ifindex for precisely this reason. If we have a 24 bit
ifindex for MPLS, this would be yet another table/hash/hack to
map between 32 bits and 24 bits. (Yes, I know there are
efficient hacks that can be done, but there is some interesting
properties that are associated with ifindices in SNMP like they
should not repeat, strict rules for deleting them which make
this somewhat more undesirable IMHO)

Of course if we increase the index to 32 bits then we need to
have another word to play with since we will at least need one
flag to distinguish between IP addresses and ifindices.


Bora

>>>>> "Bora" == Bora Akyol <akyol@pluris.com> writes:

    Bora> Curtis I was under the impression that ifindex was
    Bora> agreed to be 32 bits long based on the discussion
    Bora> about 2-3 months ago on the MPLS mailing list. Is
    Bora> there going to be an ID that clarifies this in a
    Bora> consistent manner for all of us?

    Bora> Bora


>>>>> "Curtis" == Curtis Villamizar <curtis@workhorse.fictitious.org> writes:

    Curtis> Tony, Henk,

    Curtis> Please change the following:

    Curtis> If the interface being advertised for Traffic
    Curtis> Engineering purposes is unnumbered, the first two
    Curtis> octets of the IPv4 neighbor address sub-TLV are set
    Curtis> to zero and the next two octets are set to the
    Curtis> interface ID of the unnumbered interface. In
    Curtis> combination with the IPv4 interface address sub-TLV
    Curtis> this identifies the unnumbered link over which the
    Curtis> advertised adjacency has been established.

    Curtis> Please specify that the first octet to be set to
    Curtis> zero and the remaining three octets contain the
    Curtis> ifindex.  This will be consistent with the
    Curtis> specification and the practice in OSPF (yes, I know
    Curtis> its an ISIS doc) and it will also be much more
    Curtis> convenient for big routers with lots of interfaces
    Curtis> that number their pseudo-interfaces with three
    Curtis> significant octets of numbering (like ours).

    Curtis> Curtis
    Curtis> _______________________________________________
    Curtis> Isis-wg mailing list - Isis-wg@external.juniper.net
    Curtis> http://external.juniper.net/mailman/listinfo/isis-wg

    Bora> _______________________________________________
    Bora> Isis-wg mailing list - Isis-wg@external.juniper.net
    Bora> http://external.juniper.net/mailman/listinfo/isis-wg