Hi, Consider uA advertisement(these examples are from the draft, similar line of argument applies to uDt as well) 2001:db8:0:0N00:FNAJ
Generally, it is optimal to download following route, as it avoids double lookup/packet-re-circulation within the same node. 2001:db8:0:0N00:FNAJ::/80(This route has both node and function components in it) However, the draft also recommends downloading following route: 2001:db8:0:FNAJ::/64(This route does NOT have node component in it) I guess this route is required if during compression, an uA SID gets split into two separate micro-SID-containers Say, the DA has: 2001:db8:0:A00:B00:C00:D00:N00 SRH has 2001:db8:0:FNAJ:: Or, maybe for a packet is sent from a controller, maybe it wants to sent just the uA SID for the first-hop router 2001:db8:0:FNAJ:M00: However, we think both these use-cases are not important enough. Also, the advertisement of uA/uDt are inclusive of the node-part, (i.e. uA SID is 2001:db8:0:0N00:FNAJ and not just 2001:db8:0:FNAJ) it doesn't make sense for a compressing router to split the SID itself onto two different containers. One more very strong case, not to allow scenarios that might require a router to program this route: 2001:db8:0:FNAJ::/64 If this is disallowed, and we always program uA/uDt SIDs along with the node-part, then, we can have many more local-SIDs available on a NODE. Instead of LIB space being per block, it becomes per-locator. If we use this method, a router can just use one more locator from the same block and start advertising local SIDs from that locator: For e.g. 2001:db8:0N00:FNAJ:: And 2001:db8:0N01:FNAJ:: Can be advertised from the same router. This has the great multiplier effect of number of local locators on a node.(If a node uses two different locators, the number of micro-SIDs doubles from around 8k to around 16k) Can you please share your thoughts on this? Thanks, and regards, Bharath R. Juniper Business Use Only
_______________________________________________ spring mailing list [email protected] https://www.ietf.org/mailman/listinfo/spring
