[c-nsp] ASR902 vs ME3800X
Mark Tinka
mark.tinka at seacom.mu
Tue Mar 24 07:56:03 EDT 2015
On 24/Mar/15 13:44, Adam Vitkovsky wrote:
>
>
> I would not go down that road, remember the discussion we had some
> time ago where the consensus was that single level/area is soo much
> better simpler and features friendly compared to hierarchical models.
> And anyways the main FIB consumer are the customer VPN prefixes.
Agree, but remember every node you add into the IGP creates state; and
you want to run pw's all over the place, that states need to be held in FIB.
I wouldn't say l3vpn prefixes are the majority in most MPLS networks, as
some MPLS networks focus on l2vpn's (Martini, not BGP), which would mean
less l3vpn BGP state to be held. So this one depends.
I don't like Route Leaking, but if the 4-port ASR920 does not have
enough FIB to hold my entire IGP table + some l3vpn BGP routes, it's a
non-starter. I'm not in the mood to run pw's to an "intelligent" box in
the core to instantiate services from there. Might as well run a Layer 2
network :-).
>
>
> Well you could build the L2 rings as 10GE -so you'd have 2x10GE
> cust-agg in and 2x10GE out to MPLS.
The main problem with a Layer 2 ring is protection. Easier if the ring
is terminating on the same upstream node (but still a b**ch), way harder
if the ring is terminating on different upstream nodes (a real b**ch,
but what sales droids push to customers).
Any way you cut it, Layer 2 access points that are not point-to-point
with their upstream nodes in nature are a real PITA to operate.
> If I understood our product guy well the statement from Juniper was
> that there are no slightest plans for PBB-EVPN.
> Which was a true wtf moment for me as one would think they'd like the
> box to be a competition for ASR9k.
Well that's terrible.
I'm giving EVPN two more years (like SDN) to settle down before we can
separate the wheat from the chaff. Too much handwaving at this time,
especially if you run a multi-vendor network.
To be honest, I'm not unhappy with Martini l2vpn pw's.
Mark.
More information about the cisco-nsp
mailing list