[j-nsp] MX304 reliability

[Saku Ytti]

On Wed, 6 May 2026 at 11:10, Gert Doering <gert at greenie.muc.de> wrote:

> What would be the benefit of that?  Cheaper per-port costs for the Trio?
>
> (I have no idea if the price tag on the outside of a box like the MX301
> has any real relationship to the Trio chip costs, and whether it would
> make it cheaper if Trio6.5 could serve more ports with "just one chip")

Yes, cheaper to make and simpler, so less things to break.

MX80/MX104 are such, they are 40G Trio in fabric boxes, but with 80G
of WAN ports, because ports sit on both sides.

What you pay for boxes and what they cost to make may not have strong
correlation. But of course the NPU is a big part of the cost fabric
(MX304) too.

People have bought MX80 for 3k a pop (when they were new), and I'm
sure even at 3k a pop, it's much more than BOM.


The only real downside is that the PPS budget remains same, while port
count doubles. But almost never the PPS budget matters at all, some
people think it does, because they have bad experience of some single
port per NPU design, where it wasn't linerate. But as long as your PPS
pool is shared among many ports, it becomes very unlikely to be an
issue, because in most scenarios in most ports push trivial PPS even
at 100% util, because mean packet size is like 1200B or such. And a
port or two ingesting linerate DDoS on small packets has plenty of
spare room in the PPS pool.


-- 
  ++ytti