> -----邮件原件----- > 发件人: [email protected] [mailto:[email protected]] 代表 Robert Raszuk > 发送时间: 2014年5月22日 11:10 > 收件人: Xuxiaohu > 抄送: Yakov Rekhter; [email protected]; Hannes Gredler; John G. Scudder; Alvaro > Retana (aretana) > 主题: Re: 答复: 答复: [spring] 答复: 答复: 答复: > draft-gredler-spring-mpls-05.txt as SPRING WG document > > Hi Xu, > > > If the implementation doesn't support interoperation between different > > label distribution protocols, the implementation should have separate > > LFIB entries for each distribution protocol. In this way, there are > > still multiple LFIB (MPLS2MPLS) entries for a given prefix anyway. > > How would you lookup mpls packets across those multiple tables ? One by one ?
I didn't mean multiple tables. Instead, I mean multiple entries for a given FEC within the LFIB and each is for a separate label distribution protocol. > > No matter whether R2 has another T-LDP session or not, the label > > 33 learnt from the T-LDP session with R3 should not be deemed as an > > incoming label. Incoming labels should always be those labels > > allocated by R2 itself. > > Of course. I did not draw the targetted session to say R10 from R2 allocating > 44. > The point was that if at R2 targetted LDP wins over IGP for label allocation > (and > we are talking about global per R2 label space - not per interface) labels > distributed by IGP for FEC 3.3.3.3 will not end up in LFIB. I'm talking about per-platform label space as well:) As I've said before, in the interoperation mode, the outgoing label for FEC 3.3.3.3/32 would be the label advertised by LDP. However, the incoming labels for that FEC should be the labels allocated by both LDP and IGP. In the ship-in-night mode, there are two LFIB entries for FEC 3.3.3.3/32, one is from LDP (e.g., 44 (in)->33(out)) and the other is from IGP (e.g., 1033 (in)->1033(out)). As for global labels, it just mean all LSRs allocate the same label for a given FEC. > The same actually for labeled-bgp and IGP global labels. > > So I guess we understand each other's point now. You say to always install all > labels regardless of what protocol src FEC is in the RIB. > Specifically all labels locally allocated as well as global labels > distributed by IGP in > one or many LFIBs. > > I do not think this is the case today at least in few mpls implementations. It's no surprise since as I've said it's totally a implementation issue. However, IMHO, those few MPLS implementations as you mentioned above are not ideal. At least, if T-LDP is preferred to IGP, the LSR should not advertise it as a SR anymore. > Moreover I am not sure that this is always possible with the index based label > hack to accomodate overlapping label spaces across nodes. The usage of Index is just to reduce the label advertisements, and it doesn't change the nature of the MPLS architecture at all. Best regards, Xiaohu > Best, > R. _______________________________________________ spring mailing list [email protected] https://www.ietf.org/mailman/listinfo/spring
