[c-nsp] Nexus 7707 as Internet Edge Router?

[Saku Ytti]

On 31 July 2017 at 12:14,  <adamv0025 at netconsultings.com> wrote:


>> I could also give similarly BGP examples, which would make it appear like IOS-
>> XR has bad BGP compared to JunOS.
> I'd like to hear those actually.

I'm not gonna go through emails, cases and wiki, so just stuff from
top of my head.

XR re-advertises default at every update generation cycle, leading to
far-end potentially reset the route age changing their route
selection. This is against RFC4271 which forbids resending same
unchanged route. (CSCve41797)
XR ORR does not advertise any route (not even suboptimal) if next-hop
needs recursion.
Maybe 7 BGP crash bugs in 5.3.1 that hit us
Hard to track BGP ghosting issues (technically not ghosting, closing
TCP window so nothing can be communicated, while rest of the network
has converged, one box may be lagging tens of minutes or hour)
Static define on BGP process size, after which you just crash, instead
of use memory you actually have, breaking scaling

Then 'minor' stuff, like breaking compatibility with RFC1771 speakers
in minor release update (CSCva74669), unable to to talk standard
compliant RFC1771 implementation. Our product description does not
mandate that you must be 4271. Certainly does not feel good to
suddenly break bunch of BGP sessions to customers for no good reason.

Then of course the LPTS stuff, we see perhaps twice a month issue
where some XIPC TCP queue gets congested and all BGP sessions hitting
that queue get dropped. With no indication what was in the TCP queue.

I cherry-picked problems we don't have with JunOS. Of course in JunOS
you've long time had very unpredictable convergence between bgp and
krt.

Not trying to start debate about which BGP implementation is better or
worse. Just pointing out Juniper being 'worst' implementation is
highly subjective. Another network might have 0 problems with JunOS
BGP and massive issues with XR BGP.
--
  ++ytti