[c-nsp] ASR 9000 as border router

[Pshem Kowalczyk]

Hi,

On 30 December 2010 09:11, cisco-pe <cisco.pe.1977 at gmail.com> wrote:
> Hi Pshem,
> We have a core network based on CRS-1. The idea is to add an additional
> layer formed by the ASR-9K basically as concentrator/aggregator layer
> working as P.
> All the PE will be connected to the ASR-9K. We don´t want to connect the PE
> directly to the original core (CRS-1).
> What is your experience working with the ASR-9K in P/PE mode?

So far - pretty good. We're running IOX 3.9.1 and encountered a few
limitations of that particular software version (automatic backup
paths for TE, auto-mesh of tunnels) but they're getting added as we
speak. It looks like the platform has been build around ethernet
aggregation and generally ethernet services. We run it only on
ethernet (core/backbone links - 10G, access/aggregation 1G and 10G) so
I can't comment about the POS capabilities. Hardware-wise we have not
run into any serious problems. When comes to ethernet services it
offers quite a lot - we use it for providing L3/L2 VPNs (mainly VPLS)
with QoS. Things like CFM or Y.1731 are there already, when it comes
to 'configurability' we haven't encountered anything that would
seriously limit our deployment.
The biggest issues we found are around running BFD over Ether-bundles,
but that's getting fixed as well. With QoS over Ether-bundles - some
limitations apply when comes to shaping/policing, but nothing that
can't be worked around.

> What is your oppinion about our design, advantages/disadvantages?

We'll be deploying dedicated P layer on top of the existing P/PE one,
mainly to give us a little bit 'low-touch' core, so the 9k will run as
pure Ps. So far we came up with the following limitations (that also
might apply to your situation):
- more devices on the network (line cards/sfps etc) - more things that
can go wrong
- the uplinks between the PEs and the 'aggregation' Ps and then onto
your CRS will have to be properly dimensioned for the traffic, also in
fault scenarios
- the TE will become a little bit more complicated (if you run it)
- to 'upscale' particular link you'll have to 'upscale' at least two
links now (PE <-> ASR9k and ASR9k <-> CRS1).

I'd probably add interim layer if there was enough traffic that was
pushed between PEs hanging of the same 9k. If most of the traffic will
flow back to the CRS layer then the advantages will be smaller and
cost of upgrades higher.
If the scale of your deployment is significant when comes to the
number of nodes the extra layer provides advantage of  being able to
'abstract' out particular clusters of PEs and 9k-Ps and dedicate them
to particular purpose.

kind regards
Pshem