12 bit field would be more than enough plus it maps to all the schemas for 802.1q based topology separation. This started out as an enterprise networking discussion and it would be at the right scale for anyone in that business without complaints and an easy way to scale with 2547, EVPN or even Loc/Id (LISP, ILA, etc) methods. It may be an interesting question for people thinking of RIFT in DC microservice use cases. And whether there is even an opportunity to look at integration framework between those.
Yan On Thu, Apr 12, 2018 at 12:04 PM, Tony Przygienda <tonysi...@gmail.com> wrote: > OK, I read too fast then ;-) The opaque is what we call TID then > (topology ID, I observe that we ran out of good 3 letter acronyms years ago > ;-) > > agreed with all ... > > 0 for default topology helps obviously because you say "0 on send, ignore > on receive" on a good spec so even if the other side is not aware of the > extension, good chance they'll send you 0 which you interpret as "hey, > default stuff" and life goes on in default topology > > --- tony > > On Thu, Apr 12, 2018 at 8:49 AM, Pascal Thubert (pthubert) < > pthub...@cisco.com> wrote: > >> Hello Tony >> >> >> >> I agree that 0 is default, this helps for backward compatibility as well.. >> >> Note that the field is not the TID (T is for transaction). I’m proposing >> to add a new Opaque field since what is carried is opaque to ND. >> >> New text would say: >> >> >> >> A new Opaque field is introduced to carry opaque information in >> case the >> >> registration is relayed to another process, e.g.; injected in a >> routing >> >> protocol. >> >> A new "I" field provides an abstract type for the opaque >> information, and >> >> from which the 6LN derives to which other process the opaque is >> expected >> >> to be passed. >> >> A value of Zero for I indicates an abstract topological >> information to >> >> be passed to a routing process if the registration is >> redistributed. >> >> In that case, a value of Zero for the Opaque field is >> backward-compatible >> >> with the reserved fields that are overloaded, and the meaning is >> to use >> >> the default topology. >> >> >> >> 8 bits is what’s left in the option that we need to keep backwards >> compatible. >> >> >> >> 0 1 2 3 >> >> 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 >> >> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ >> >> | Type | Length | Status | Opaque | >> >> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ >> >> | Rsvd | I |R|T| TID | Registration Lifetime | >> >> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ >> >> | | >> >> ... Registration Ownership Verifier ... >> >> | | >> >> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ >> >> >> >> What do you think? >> >> >> >> Pascal >> >> >> >> *From:* Tony Przygienda <tonysi...@gmail.com> >> *Sent:* jeudi 12 avril 2018 17:43 >> *To:* Pascal Thubert (pthubert) <pthub...@cisco.com> >> *Cc:* 6lo@ietf.org; Yan Filyurin <yanf...@gmail.com>; >> draft-ietf-6lo-rfc6775-upd...@ietf.org; r...@ietf.org >> *Subject:* Re: instance ID in rfc6775 update >> >> >> >> Yes, we do have discussions over RIFT where it seems a multi-plane or if >> you want multi-topology concept as introduced originally in RFC5120 would >> be helpful. RIFT can be very easily instantiated on multiple ports and with >> that has no problem to run multi-instance/topology but the dataplane >> correlation from the leaf would be very helpful. RIFT leaf implementation >> is very "thin" and with that architectures that don't rely on either LOC-ID >> or BGP overlays become feasible, albeit obviously not @ the scale something >> like 2547bis or EVPN can operate. >> >> So in short, I think I support this suggestion fully. >> >> >> For the practical encoding, I suggest to choose TID=0 as "default >> topology", i.e. "what you do today" and avoid an I bit which will cause an >> encoding corner case if it's not set but TID<>0? >> >> From experience, 8 bits is just about enough but 12 bits are plenty for # >> of topologies people sometimes think they need on building network >> architectures ... >> >> >> >> my 2c ... >> >> >> >> --- tony >> >> >> >> On Thu, Apr 12, 2018 at 7:23 AM, Pascal Thubert (pthubert) < >> pthub...@cisco.com> wrote: >> >> Hi again >> >> >> >> A proposed text would be like: >> >> >> >> >> >> 0 1 2 3 >> >> 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 >> >> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ >> >> | Type | Length | Status | Opaque | >> >> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ >> >> | Rsvd | I |R|T| TID | Registration Lifetime | >> >> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ >> >> | | >> >> ... Registration Ownership Verifier ... >> >> | | >> >> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ >> >> >> >> …. >> >> >> >> Opaque: >> >> One-byte Opaque field; this is an octet that ND does not need to >> process >> >> but that the 6LN wishes the 6LR to pass transparently to another >> process. >> >> I: >> >> Two-bit Integer: A value of zero indicates that the Opaque field >> carries >> >> an abstract index that is used to decide in which routing topology >> the >> >> address is expected to be injected. In that case, the Opaque field >> is >> >> passed to a routing process with the indication that this is a >> topology >> >> information and the value of 0 indicates default. All other values >> are >> >> reserved. >> >> >> >> >> >> Does that work? >> >> >> >> Pascal >> >> >> >> *From:* Pascal Thubert (pthubert) >> *Sent:* jeudi 12 avril 2018 15:40 >> *To:* 6lo@ietf.org >> *Cc:* Yan Filyurin <yanf...@gmail.com>; Tony Przygienda < >> tonysi...@gmail.com>; draft-ietf-6lo-rfc6775-upd...@ietf.org >> *Subject:* instance ID in rfc6775 update >> >> >> >> Dear all : >> >> >> >> During a conversation on the RIFT protocol it appeared that there are use >> cases in RIFT to support host mobility with rfc6775-update. >> >> There is a caveat, though, which is in fact common with RPL. Both cases >> need a concept of multi topology routing. >> >> In the case of RPL, the topology is indexed by an instance ID. In the >> case of RIFT, there is a need for an index to a RIB, so one octet is >> probably enough. >> >> A suggestion is thus to use the reserved octet in the ARO to carry an >> instance ID, and use a bit to signal that this is what that field does, in >> case there is a need later to overload it with something else. >> >> >> >> I understand this is coming late in the process; but then there is no >> logic associated to the change, this is just passing on an additional >> information that is useful for more than one candidate protocol. >> >> >> >> Please let me know if there is an issue pursuing this. If there is no >> opposition, my plan it currently to add this in rev-19. >> >> >> >> All the best, >> >> >> >> Pascal >> >> >> > >
_______________________________________________ 6lo mailing list 6lo@ietf.org https://www.ietf.org/mailman/listinfo/6lo
