[10:05:39] --- ignas has joined
[10:29:13] --- stpeter has joined
[10:29:25] --- stpeter has left
[10:35:24] --- ignas has left: Replaced by new connection
[10:51:30] --- ignas has joined
[12:18:56] --- ignas has left: Replaced by new connection
[12:24:12] --- ignas has joined
[12:26:11] --- dinakar has joined
[12:26:14] --- dinakar has left
[12:58:49] --- ignas has left: Disconnected
[12:59:38] --- ignas has joined
[13:01:30] --- tony1athome has joined
[13:02:19] <tony1athome> Is there a scribe here?
[13:05:35] <tony1athome> I guess not. For the sake of posterity, I'll take some notes.
[13:05:53] <tony1athome> Alex has already covered the status of documents.
[13:06:24] <tony1athome> One of the documents is about FRR, and this draft is being discussed.
[13:06:53] <tony1athome> Specifically, the concern is about loops created during FRR computation when the path crosses an area boundary.
[13:09:01] <tony1athome> The document is draft-atlas-rtgwg-ipfrr-spec-base-03.txt
[13:09:36] <tony1athome> There is an issue that comes up with domain boundaries
[13:10:01] <tony1athome> Correction, it's about AS external routes (presumably in OSPF)
[13:11:09] <tony1athome> DWard is pointing out that this is a fairly involved situation
[13:11:24] <tony1athome> Now reporting on micro-loop DR results
[13:11:59] <tony1athome> Dropping incremental cost changes, synchronized FIB installation
[13:13:31] <tony1athome> Synchronized FIBs would require NTP synchronization, which was viewed as not desireable
[13:13:56] <tony1athome> draft-zinin-microloop-analysis-00.txt
[13:17:01] <tony1athome> Other alternatives: tunnels, ordered FIB updates
[13:18:10] <tony1athome> DT to make recommendations
[13:18:33] <tony1athome> PLSN looks like most likely proposal
[13:20:03] <tony1athome> Dward: Why is imperfection good enough?
[13:20:22] <tony1athome> Zinin: Price performance not worthwhile
[13:27:21] <tony1athome> Next proposal: notvia addresses
[13:27:52] <tony1athome> Solves all problems
[13:28:41] <tony1athome> Specify address on other side of failure, tunnel packets
[13:29:11] <tony1athome> Need address on other side of failure, how?
[13:30:19] <tony1athome> Do an SPF to all 'notvia' addresses. Compute SPF per failure
[13:30:27] <tony1athome> Incremental SPF with early termination
[13:31:02] <tony1athome> Takes between 5 and 13 full SPF times
[13:31:41] <tony1athome> Distribute 'notvia' in LDP for MPLS
[13:32:12] <tony1athome> [example]
[13:34:50] <tony1athome> Now covering complexities in problem
[13:34:55] <tony1athome> What about LAN failures?
[13:35:51] <tony1athome> Doing BFD with the switch helps isolate the problem
[13:36:04] <tony1athome> Mcast is hard [not news ;-) -- tli]
[13:36:46] <tony1athome> Needs encapsulation
[13:36:57] <tony1athome> ECMP also needs encapsulation
[13:37:07] <tony1athome> Incremental deployment
[13:37:26] <tony1athome> Capable routers advertise 'notvia' capability
[13:37:41] <tony1athome> Avoid routers not advertising
[13:38:12] <tony1athome> Extensions to protocols necessary
[13:38:39] <tony1athome> Add one 'notvia' address per failing component
[13:39:03] <tony1athome> Need to advertise type of encapsulation
[13:40:02] <tony1athome> Focus on node failures
[13:40:09] <tony1athome> Link failures fall out for free
[13:40:49] <tony1athome> Multi-homed prefixes
[13:43:17] <tony1athome> [Messy and I'm not following, sorry -- tli]
[13:44:44] <tony1athome> Summary: This seems to work well, covers all sorts of related complications
[13:45:25] <tony1athome> How many 'notvia' addresses need to be advertised?
[13:45:33] <tony1athome> One per neighbor per node
[13:47:19] <tony1athome> What happens if we're doing incremental deployment
[13:47:25] <tony1athome> No worse than today
[14:08:44] <tony1athome> More discussions about tradeoffs
[14:14:26] --- dr has joined
[14:32:45] --- tony1athome has left
[15:01:32] --- dr has left
[16:00:17] --- ignas has left: Replaced by new connection
[16:00:17] --- ignas has joined
[16:00:18] --- ignas has left
[23:39:51] --- lixia has joined
[23:40:53] --- lixia has left