[j-nsp] under the hood of MP-BGP

adamv0025 at netconsultings.com adamv0025 at netconsultings.com
Tue Jan 3 13:30:09 EST 2017

> > Jeff Haas [mailto:jhaas at juniper.net]
> > Sent: Tuesday, January 03, 2017 3:18 PM
> >
> > This step is actually why I posted the question. We encountered a
> > problem where it takes 5-10 minutes for local routes, that are already
> > installed in other VRFs on the PE, to get installed from L3VPN table
> > into a newly configured VRF table.
> >
> > By this, you mean you have some route that has a route-target that is
> > already installed in VRF A to be installed also in VRF B that has an
> overlapping target?
> >
Yes please:
route x.x.x.x:RT1:1 (mgmt. prefix)
Existing VRF import policy RT1:1
And adding NEW VRF import policy RT1:1
> >
> > Since I posted this question we've started testing the behaviour and
> > in the lab it seems that the "local routes parsing/import" is done as
> > the first thing, or at least right after the route refresh is sent
> > out, so the routes are leaked from bgp.l3vpn.0. into the NEW VRF table
> right away in the lab.
> >
> > The work is done in parallel.  Think of it as the same type of work
> > that is done when policy needs to be re-evaluated; we will walk the
> > tables as part of the work until it's done.
> >
Yes please, that's what I thought, simple top to bottom walk of the L3VPN
table to see if any of the routes needs to be installed into the newly
configured VRF.
> > So it must be something in the production (config) that is slowing
> > things down.
> >
> > That usually implies scale.  Remember the work is part of general
> > table re- evaluation, so it's likely the total number of routes in your
> >
But for 1M destinations/routes & 1000 existing VRFs we tested with it was
just couple of (~4) seconds to get the routes into the VRF under test. 
So in production with 1M destinations and 2M routes it should be around 8
seconds then but for some reason it's 5-10minutes. 
I'm thinking whether it's some odd import policy on one of the VRFs in
production or something that can be holding up or halting the L3VPN table
walk and import thread (cause VRFs in the test bed are pretty uniform).
Thank you very much

::carrier-class solutions for the telecommunications industry::

More information about the juniper-nsp mailing list