Hi, Robert: Aijun Wang China Telecom
> On Nov 27, 2021, at 21:59, Robert Raszuk <[email protected]> wrote: > > > Hi Aijun, >>> >>> Yes BGP option 2 as I described gives you PIC effect. In fact I am quite >>> convinced that if done right it can be much faster then IGP flooding >>> especially at scale. Please recall all safety belts build into IGP to slow >>> down churn when multiple events happens in a time scale. That will have a >>> negative effect on PUA/PULSE propagation speed at scale. >> [WAJ] If such thing happen, won’t the BGP update will also be affected? The >> underlying IGP are converging. > > You need to be able to distinguish local area convergence from hierarchical > domain wide churn. No BGP will not be affected at all neither in the local > area nor in any other area. [WAJ] The PUA/PULSE message doesn’t trigger the SPF calculation, how can the domain wide churn happen? > > >> [WAJ] When you recall the NH prefix, the 32/ or 128/ route is not exist >> within the RIB of receiving node, only the summary routes from IGP is >> existing. How to prefer the 32/ or /128 route? > > In Option-2 I suggested it does exist. [WAJ] Recall but still exists? Which vendor has such knob? Please give some references > > >> [WAJ] Even with this non existing assumption, you will require the ABR to be >> configured with other PE for BGP sessions. Such design will let each ABR act >> again as RR, won’t you think it is redundant? > > No. What you have envisioned would be a pretty ugly design. In Option-3 I > proposed ABR will just have 2 sessions to area-0/level-2 RRs (current or > new). > > >> [WAJ] Register with Loc RIB is easy, but register with other node within IGP >> is not easy. There is currently no such mechanism within IGP, as Tony Li >> also confirmed. > > Well there is at least one registration mechanism today with BGP where PE > registers with RRs on what is interesting for it. Check out RFC4684. [WAJ]Just image your proposal in the whole picture: should every PE establish two BGP sessions with the RR? May the P router should also do so? And then every PE router configure it is interested all the loopack addresses of other PEs? Maybe all the P routers needs to do so? Is it efficient than the IGP flooding? > > >> [WAJ] The tunneled traffic will be diverted to other backup tunnel until the >> primary node is back again. The detection of the UP status of the primary >> node is depend on the tunnel technology itself. PUA/PULSE just notifies the >> tunnel endpoint is down. > > Encapsulation does not require tunnel setup. Hint: MPLS over UDP. Show me > please where is the "tunnel" ? [WAJ] We are discussing tunnel, why the encapsulation jumps in? > > >> [WAJ] Using the backup tunnel until the primary node is brought up again. > > :) > > Thx, > R.
_______________________________________________ Lsr mailing list [email protected] https://www.ietf.org/mailman/listinfo/lsr
