Hi Acee, RFC 5185 says that interface data structure is created for each multi-area adjacency. I guess that we are not allowed to allocate several ifIndex values for the same IP interface, because it is property of router's interface, not OSPF interface. Hence, we have several OSPF interfaces with the same ifIndex in unnumbered case and, thus, ambiguity in Interface table. The same for numbered - we have IP interface address (one), which is the same for multiple OSPF interfaces, and we again obtain ambiguity. Per my understanding advertising neighbor's IP address (or ifIndex) in Link Data doesn't help here.
> 30 нояб. 2020 г., в 20:20, Van De Velde, Gunter (Nokia - BE/Antwerp) > <[email protected]> написал(а): > > The oddnes is that the architecture decision in RFC5185 to select > remote-ip-address instead of local-ip-address for the ‘Link Data’ is making > things much more complicated. > I am surprised to see that using the remote-ip-address is seen as the > ‘better’ choice as selecting local-ip-address. To me it seems as a worse > choice. > > A question that was asked: How router will be able to match Link TLV (RFC > 3630) to corresponding Link in Router LSA? > > Answer: > For unnumbered links we can match Link TLV with Router TLV using the IfIndex > when there is no stub type 3 link (=easy) > For numbered: > we must first look if the there is a stub type 3 link > If stub type 3 exists, then the RFC3630 local ip address must be used to > identify the correspond link within the router TLV to the neighbor > If the stub type 3 link did not exist in Router TLV link, then the maybe the > link is unnumbered, and we try to match upon IfIndex… This may give a match > or no match > If there is no match, then maybe the link is MADJ and we must use the RFC3630 > remote IP address to match upon the Link Data > = over-complex. (If we used for RFC5185 ‘Link Data = local ip address’ then > (2) would given answer directly) > > In addition, for a router it is much simpler to learn and advertise > local-ip-address in Router LSAs then using a remote-ip-address. > I also believe that if we want to search something in TEDB after receiving a > TE Link TLV. How can we identify from the TE Link TLV if multi-area or not > multi-area? If we can not, then how can we create the correct key? > > Looking at the above, the choice of using remote-ip-address for RFC5185 Link > Data seems not the best design that we can do, and is adding OSPF complexity > without benefits. > > Should this not be corrected in RFC5185 and simply use local-ip-address > instead of the remote-ip-address for Multi-area Link Data and avoid the > additional unnecessary complexity the current RFC for numbered links? > > Brgds, > G/ > > > From: Lsr <[email protected]> On Behalf Of Acee Lindem (acee) > Sent: Monday, November 30, 2020 18:01 > To: Alexander Okonnikov <[email protected]>; Peter Psenak > (ppsenak) <[email protected]> > Cc: [email protected] > Subject: Re: [Lsr] Link Data value for Multi-area links > > Hi Alex, > > Multi-Area interface disambiguation is required to support the OSPF MIB as > specified in RFC 4750. The table indexing doesn’t include the area. For > example: > > -- OSPF Interface Table > > ospfIfTable OBJECT-TYPE > SYNTAX SEQUENCE OF OspfIfEntry > MAX-ACCESS not-accessible > STATUS current > DESCRIPTION > "The OSPF Interface Table describes the interfaces > from the viewpoint of OSPF. > It augments the ipAddrTable with OSPF specific information." > REFERENCE > "OSPF Version 2, Appendix C.3 Router interface > parameters" > ::= { ospf 7 } > > ospfIfEntry OBJECT-TYPE > SYNTAX OspfIfEntry > MAX-ACCESS not-accessible > STATUS current > DESCRIPTION > "The OSPF interface entry describes one interface > from the viewpoint of OSPF. > > Information in this table is persistent and when this object > is written the entity SHOULD save the change to non-volatile > storage." > INDEX { ospfIfIpAddress, ospfAddressLessIf } > ::= { ospfIfTable 1 } > > Note that if you really want to support this optimally, you could use a > separate subnet pre-area and have adjacencies on secondary addresses. My > Redback/Ericsson implementation allowed for this. > > Thanks, > Acee > > > From: Lsr <[email protected] <mailto:[email protected]>> on behalf of > Alexander Okonnikov <[email protected] > <mailto:[email protected]>> > Date: Monday, November 30, 2020 at 5:27 AM > To: "Peter Psenak (ppsenak)" <[email protected] <mailto:[email protected]>> > Cc: "[email protected] <mailto:[email protected]>" <[email protected] <mailto:[email protected]>> > Subject: Re: [Lsr] Link Data value for Multi-area links > > Hi Peter, > > > 30 нояб. 2020 г., в 12:56, Peter Psenak <[email protected] > <mailto:[email protected]>> написал(а): > > Hi Alex, > > On 27/11/2020 13:49, Alexander Okonnikov wrote: > > Hi Peter, > Which kind of ambiguity is meant? In case of numbered point-to-point each > link has its own unique IP address, so there is no ambiguity. > Per my understanding this problem has appeared due to follow reasons: > 1) In old versions of the draft (up to -05) it was proposed that multi-area > links are treated as unnumbered. ifIndex to be encoded in Link Data field, > irrespectively whether interface has its own IP address (numbered) or borrow > it (unnumbered); > 2) From -06 to -08 multi-area links are still treated as unnumbered, but if > interface is numbered, then IP address of the neighbor (rather than local > one) to be encoded into Link Data, in order to make the link look like > unnumbered; > 3) In version -09 of the draft and in RFC 5185 itself there is no more > mentions that multi-area link to be treated as unnumbered. Rather, another > approach is used - if router's interface is numbered, then link is also > numbered; if router's interface is unnumbered, then link is unnumbered. The > rule that specifies omitting corresponding type 3 link is added. Mention of > 'unnumbered' link is also removed from section 3 in RFC 5185. > > Hence, in version -09 with removing unnumbered nature of multi-area links > Link Data for numbered links had to be changed from Neighbor's IP address to > own IP address, as it is specified in RFC 2328. From perspective of other > routers this link can be treated as numbered or unnumbered, depending on > configuration of neighbor's corresponding interface. > > you are free to advertise the link as unnumbered. RFC5185 is not mandating to > send IP address really. > > The same valid for numbered ones. I.e. I'm free to advertise the link as > numbered. This is straightforward when the link is numbered indeed. And if we > would prefer to have deal with unnumbered interfaces, we would not need RFC > 5185 (section 1.2). > > > One question - how neighboring router will perform next-hop calculation (in > case it needs to do so)? If neighbor is configured with numbered interface, > it will treat Link Data as IP next hop, which will be its own IP interface > address. > Another question - how router will be able to match Link TLV (RFC 3630) to > corresponding Link in Router LSA? For example, we want to calculate RSVP-TE > LSP based on IGP metric (RFC 3785) and thus router needs to match IGP link to > TE link. > > I don't believe you are going to do any traffic engineering over a multi-area > adjacency. MADJ is there to address the OSPF route preference rules that may > lead to sub-optimal routing. MADJ link is not advertised for TE purposes. > > Why not? We need multi-area configuration and at the same time we need > ability to build intra-area RSVP-TE LSPs within each of areas. And what about > calculating IP next hop? Which compatibility is meant in section 3? > > thanks, > Peter > > Thank you. > > > Thank you. > > 27 нояб. 2020 г., в 14:50, Peter Psenak <[email protected] > <mailto:[email protected]>> написал(а): > > Alexander, > > On 26/11/2020 17:58, Alexander Okonnikov wrote: > > Hi WG, > RFC 5185 says that Neighbor's IP address to be encoded into Link Data field. > Per RFC 2328 router's own IP address to be encoded into Link Data. What is > the reason to advertise neighbor's IP address for multi-area links and not > local IP address? It seems like bug. Could someone comment on this? > > Advertising a neighbor address/ifindex helps to eliminate ambiguity in case > of parallel point-to-point adjacencies. It's not perfect, but that's how it > was specified. So it's not a bug. > > thanks, > Peter > > > Thanks in advance. > _______________________________________________ > Lsr mailing list > [email protected] <mailto:[email protected]> > https://www.ietf.org/mailman/listinfo/lsr > <https://www.ietf.org/mailman/listinfo/lsr>
_______________________________________________ Lsr mailing list [email protected] https://www.ietf.org/mailman/listinfo/lsr
