Hi Hannes, ________________________________________ 发件人: Hannes Gredler [[email protected]] 发送时间: 2013年11月8日 7:24 收件人: Xuxiaohu 抄送: [email protected]; [email protected] 主题: Re: [OSPF] Inconsistency between OSPF extention and IS-IS extension for segment routing
hi xuxiaohu, On Nov 7, 2013, at 11:51 PM, Xuxiaohu wrote: > OSPF extension draft proposes to use Extended Prefix Opaque LSA to carry > SR-related attributes. Since the Extended Prefix Opaque LSA does not > advertise reachability of the prefix, but only its attributes, the prefixes > contained within those LSAs for building IP routing table (e.g., Router LSAs) > can be aggregated when crossing area boundaries while the Extended Prefix > Opaque LSAs containing prefix SIDs can be intactly propagated across area > boudaries. The final effect is much similar to the mechanism defined in > RFC5283. > > In contract, IS-IS extension draft proposes to reuse those Extended IP > Reachability TLVs which are used for building IP routing table to carry > SR-related attributes. Although this choice has the benefit of propagating > less LSAs, it loses the capability of aggregating routes when a crossing > level boudaries. Furthermore, it requires the L1/L2 routers much be > SR-capable. and why is it a problem squeezing a few thousand transport loopbacks across an ABR / L1L2 router ? [Xiaohu] I should have said that if you want to aggregate those /32 prefixes for transport loopback addresses when crossing L1/L2 routers, those L1/L2 routers MUST be SR-capable. Otherwise, those prefixes can not be aggregated. The reason is that it requires L1/L2 routers to translate a TLV containing prefix SID into two TLVs, one is the TLV containing the aggregated route, the other is the TLV containing the prefix-SID. In addition, it seems that there is no mention about whether the /32 prefix contained in the latter TLV will be still used for building IP routing table. If yes, these host routes are actually leaked across levels. If no, the Metric field SHOULD be set to a value larger than MAX_PATH_METRIC (i.e., 0xFE000000). > Although these two drafts are proposing extensions to two different IGPs, > IMHO, it would better to provide similar capabilities if possible, especially > advoid destroying the existing capabilities of these two IGPs, e.g., > inter-area/level route aggregation capability. most SPs running hierarchical IS-IS have turned on domain wide prefix leaking. in fact doing aggregation in those environments is prohibitive in determining the true cost to a particular BGP next hop. - so we do not really loose much. [Xiaohu] Did you mean that RFC5283 is useless in most SP networks? > To Peter Psenak, > > I don't agree with your argrment that the reason that IS-IS extension draft > made that choice is because there is no choice for IS-IS. i second that opinion - the only extensible container for IP prefixes are the extended IP reach TLVs. > In fact, you can use the signalling mechanism for Label Request which has > been proposed in draft-xu-rtgwg-global-label-adv-00. That's to say, you can > use separate Extended IP Reachability TLVs other than those for IP > reachability advertisement to carry SR-related attibutes. Since the former > TLVs are intened for advertising label bindings other than building IP > routing table, the Metric field of these TLVs is set to a value larger than > MAX_PATH_METRIC (i.e., 0xFE000000). It's a normal approach for IS-IS. Of > course, if SR is just used within a single level, it's good to use the > existing approach proposed in the IS-IS extension draft. keep in mind that we do not advertise a label per prefix but rather a global *index*. it is my understanding that the semantics of draft-xu-rtgwg-global-label-adv-00 are quite different in the sense that absolute label values are being advertised here. [Xiaohu] Thank for pointing out this. In fact, if the segment routing paradigm finally uses global indexs, rather than global labels, the mechanism defined in the above draft can be easily adapted to advertise global indexs as well. BR, Xiaohu /hannes _______________________________________________ OSPF mailing list [email protected] https://www.ietf.org/mailman/listinfo/ospf
