IETF Routing WG Minutes November 4th, 2013 Georgia B - 13:00 - 15:00 Chairs: Alia Atlas and Alvaro Retana Scribe: Acee Lindem - acee.lindem@ericsson.com Agenda Bashing: Arun will not present Dynamic Path Selection Document Status (See Slides): - "Advanced Multi-Path Requirements" - WG Last Call Completed and publication requested. - "MRT Algorithms" - Will be adopted pending IPR responses. Advanced Multi-Path Use Cases - Curtis Villamizar (See Slides) - Terminology change from "Composite Link" to "Advanced Multi-path". - Includes two foundation use cases. - Will go to WG last call soon. Remote LFA - Stewart Bryant (See slides) - See changes to current version in slides. - Steward does not want to merge with the node repair document. There was consensus in the room to move on w/o merging. - Extended P-space vs P space explained. Consensus is that the example algorithm should include both the P-space and Extended P-space. Alia: Believes there is still confusion on P-space vs Extended P-space. Chris Bowers: Confusion with all the spaces in the document. Can it be simplified? Stewart: I can move Extended P-space ahead of Q-space in description. - Cost-Based Algorithm - Question of style and will not be changed. Uma Chunduri - Why is algorithm not in appendix? Stewart - Graph vs Cost Reachability algorithm - both algorithms put into the main body. Uma - Prefers appendix. Chris - The text says it is just an example. Uma - The downstream check is in the algorithm but not the text. Stewart - It was in the algorithm provided by Chris. Alia - RFC 5286 indicates that downstream checking can be useful in avoiding loops. - LFA with RLFA versus only R-LFA Chris: The section with the alternate view creates confusion. Email from Chris: Following up on an item discussed in the meeting. We agreed that the paragraph in section 4.2.2 that describes extended P-space in terms of (local)LFA should be removed because this alternative description may confuse the reader more than it enlightens Stewart: Wants LFA to be the primary with RLFA added. Acee: LFA has advantage of providing Extended P-Space and providing protection w/o tunneling. Hannes: Agrees with Acee and clarifies Stephane's comment. Uma: Agree with making LFA primary calculation. Stewart: Will add text on policy engine to decide on LFA and RFLA calculation applicability. Stewart: Points out that you can calculate both LFA and RLFA. - Targeted LDP Address - Advertise in protocol. This is preferred. However, some boxes don't advertise. - Management configuration as a requirement. - Pick an address arbitrarily. Uma: What configuration is expected? Stewart: Could be silent since it is a vendor implementation issue. Hannes: Lowest IP address is useless. A node should be able to advertise that it doesn't support T-LDP. Rob Shakir - Draft should not remain silent on this issue. Stewart: Will use Router-ID unless configured to do otherwise, e.g., using a different address. Uma: Do coverage numbers in draft include downstream protection? Stewart: No - will include numbers if they can be readily obtained. - Figure 1 changed or not? Stewart: Will not be changed. - Final Comments? Hannes: I don't understand reluctance to merge with node protection. Stewart: Does not want to merge. Wants to finish draft. Node protection problem can quickly get out of control. Hannes: This would be valuable implementation experience. Alvaro: Already have "rough" consensus on moving forward with the draft without merging with node protection. Algorithm for Ordered Metric Adjustment - Gang Yan (See slides) - Background cites previous work (see slides) - Main different with RFC 5715 is performance - only two Reverse SPFs required. Pierre Francois : Length of router sequence will lengthen convergence. Huaimo Chen: Our draft does cover this issue. Pierre Francois: Disagrees. It will be far from practical in medium-sized service provider network. Huaimo: Will do more study. Alvaro: Authors should do some more analysis to address convergence and take discussion to list. Source/Destination Routing Usecases - Fred Baker (See slides) - Primary motivated by Homenet requirement. - RFC 3704 originally describes egress filtering problems in multi-homed networks. Provides manual solution. - Homenet requires automatic solution. - Need to be able to advertise and use a route to a destination from a source. Alvaro: Homenet example. Other use cases? Fred: Use case generalizes to any multi-homed campus. Stewart: Why not separate FIB for each source? Fred: What about overlapping source prefixes? Stewart: How is it done? Fred: Original OSPF draft covered this but I removed it by request. Stewart: This would be useful in this draft on use cases. Ron Bonica: What you really need is two FIBs. Ron: Other technologies handle this problem rather than inventing something new. Fred: What about overlapping source prefixes? David Lamparter - Problem with source based routing was recently fixed in Linux. Jeff Hass: Let's talk about the abstraction. Can we still view it as a multiple FIB model for simplicity? Ron: Can we not give up the multi-homed problem as multiple FIB concept without consideration? Fred: Fair enough. Alvaro - Let's have more discussion on the list regarding this draft. An Architecture of Central Controlled Interior Gateway Protocol - Robin Li (See slides) - SDN control of IGP and use cases (See slides) - Architecture with IGP controller with two models (on network device or server) Edward Crab: Has no idea what you are talking about? Is this a top down SDN controller? Huaimo: Proposes IGP as central controller. Robin: IGP as central controller - not BGP or PCEP. You can see difference. Greg Mirsky: Would be good to have a distinct section on requirements. There is not sufficient base. Is this really needed? Robin: Draft introduces architecture. Use cases to come. Based on Central Control Network. Curtis: Should start with problem statement. People don't understand what problem is being solved. Alvaro: Theme of comments - Need more discussion on list. Acee: Central controller controlling IGP instances or Central controller using IGP to control network elements?