Hi Young, et al.
I thought the draft was well written and an interesting area for discussion. I had the following general comments. a) A key goal of the work is to gather necessary network management information for offline complex path computation. Its logical to assume that the complex offline path computation might be centralized. It might be worth stating this as an assumption. b) We should not get to caught up in the details of proposed mechanisms or solutions. This is after all an architecture document. c) A hybrid approach to performing path computation in the network might be considered as there is motivation for performing path computation (simple versus complex) in different parts of the network. All planning and complex path computation could be performed by the offline and centralized PCE. Rapid restoration and local repair could then distributed to the local PCEs. Each distributed PCE can always signal the centralized PCE for reoptimization after the initial local path computation, in order to request a path computation that considers more complex constraints. d) Fundamentally we should use the right mechanism to distribute the data according to where it is needed. If some form of path computation is needed on LSRs, then in my opinion the necessary information should be distributed using the IGPs. We should be careful about recommending a solution that does not use the IGP at all. In a model where there are many nodes with path computation capabilities, we effectively need to flood information to each PCE. The IGPs have been designed for this. If we use PCEP to distribute information to go in the TED we may end up trying to turn PCEP into a flooding protocol that replaces the IGPs. e) As mentioned in point (b), we should avoid getting into solution details. As an architecture/requirements document, this draft should maybe not mention any protocols at all except in the context of what they already do and are used for. So, discussing using PCEP for data distribution, or discussing LDAP may be out of scope (even for an Appendix). But it is very valid to describe the data model, and I think that bit that is not generally flooded is a distributed database just like in the LDAP model. f) I made some minor comments and suggestions along the points above as change modifications in a word document. I will unicast the document to you shortly. Br, Dan.
_______________________________________________ Pce mailing list [email protected] https://www.ietf.org/mailman/listinfo/pce
