[c-nsp] Typhoon support on XRe

[adamv0025 at netconsultings.com]

> Mikael Abrahamsson [mailto:swmike at swm.pp.se]
> Thursday, May 04, 2017 11:39 AM
> 
> On Tue, 2 May 2017, adamv0025 at netconsultings.com wrote:
> 
> > Oh dear, 16.67Mpps budget per 10GE is now acceptable? We're doomed.
> 
> So I'd say the market has spoken, people don't care about wirespeed
> anymore, hasn't for 10 years at least, and the market is catering for
> wirespeed forwarding at around 100-200 bytes min sized packets. If you
want
> more than that, then you have to typically use fewer ports per NPU, if the
> internal wiring allows for that kind of "trickery" to get wirespeed out of
the
> NPU by using less oversubscription of its capacity. Also, with fewer
features
> the wirespeed performance might be a lot better than the worst-case
> specified by the vendor.
> 
Well people don't care about a lot of things, like how routers actually work
or whether they can still protect your EF traffic when overloaded, and
that's not necessarily a good thing even though everyone does it. 
But I guess you're right, it's just I've got a feeling we're slowly getting
ripped off here, I mean it's alright if the LC is marketed as
oversubscribed, but it's certainly not alright if a LC can deliver what it's
marketed for only under certain specific circumstances and it seems like
this "nominal performance window" is getting narrower and narrower as the BW
the LCs are marketed for is increasing. 
And you get told - use less ports then, or buy more LCs, or don't use these
two features together.   
 
> Going forward, it's not a bad thing to design your network to fit features
> commonly available in merchant silicon. People designing networks that
> require a "fully featured, classic-style core router" are going to
discover that
> there is less and less focus on them, probably meaning significantly
higher
> cost compared to the gear coming out that caters to the "less-complicated"
> networks.
> 
It's not the core routers I'm worried about it's the edge nodes -that's
where all the features are enabled and the scale comes to play. 
It's an interesting point you brought up though, talking about merchant
silicon for core routers. 
Let's consider NCS5K -it's one of the "100GE for everyone at the bar" type
of boxes, has an abundance of these dirty cheap 100GE ports, it might not
have the most feature rich NPU but it would only ever do MPLS switching so
no problem there. (Btw talking about raw performance, NCS 5001 is 35Mpps per
10GE)
But does the NPU's architecture allow it to protect your EF traffic even
when it's overloaded (under DDoS), I don't know I haven't checked/tested, I
know ASR9K LCs would. 
What I'm trying to illustrate here is that raw performance of the NPU is one
thing, but there's also other dimensions that dictate the price and
suitability of the HW for a certain project. 
For example if it's a network that can't get DDoS traffic in it and you know
you're only going to have 60% utilization before upgrade, or all traffic is
BE, then sure nothing to worry about. 
   
   

adam

netconsultings.com
::carrier-class solutions for the telecommunications industry::