[c-nsp] Cisco 8000

[James Bensley]

On Sun, 15 Dec 2019 at 06:55, Mark Tinka <mark.tinka at seacom.mu> wrote:
> I quickly notice that the NCS6000 is no longer on that list of "core
> routers".
>
> To be honest, I'm not sure whether I trust Cisco's long-term vision
> about their core routers. The CRS-X, which still has plenty of juice, is
> being abandoned, even though I think it still has at least 10 more years
> in it.

I would guess it [CRS-X] is too power and cooling inefficient. The
same for the NCS6K...

> Then it was the NCS5000 they were proposing for a decent core to
> compete against Juniper and Arista. Now I'm seeing the 8000, which uses
> their own custom silicon and is likely to contradict their message of
> pushing the NCS5000 as a core router, which is a Broadcom-based router.

The longevity of the NCS55K could be questionable although I'd expect
the 8Ks to be more expensive, I expect the Silicon One chip to be more
featureful and where the NCS55K sines is with dumb packet pushing. The
NCS55Ks are high throughput and a relativity low price point with
fewer features. They are also much more space, power and cooling
efficient compared to the NCS6K and presumably the CRS-X but I don't
know those boxes very well. Also Cisco are pumping new features in the
NCS55K right now, so I don't think it will be dropped any time soon,
it serves a different purpose and price point than the NCS8K as far as
I can see.

> Last thing I want is to find myself less than 5 years into a box only to
> be told that they aren't feeling it anymore, which is what happened when
> we picked up the CRS-X back in 2014 :-\...
>
> Meanwhile, I'm told that the NCS6000 is still available, but only for
> customers that need multi-chassis and that. I mean, in 2020, how many of
> those are still around, if not about 3 or 4?

Yeah I don't think there is any point buying NCS6K now. I think the
NCS55K has superseded it. I'm not sure there is a place for larger
ASR9K boxes either. We tried to re-purpose an ASR9922 recently that
was pulled out and it just wasn't cost effective; given the number of
packets they shift whilst consuming a whole rack of space and $LOL
amount of power, if you just need edge capacity with minimal features
and buffers it's not efficient.

Cheers,
James.