[c-nsp] MPLS and links with limited MTU size

[Saku Ytti]

On 25 January 2016 at 11:23, Adam Vitkovsky <Adam.Vitkovsky at gamma.co.uk> wrote:

> Interesting I didn't know that, do you know if LU or MQ is responsible for the fragmentation? But I'm afraid it would overload the Trio ensemble pretty quickly.
> Yeah, on ASR9K the "to be fragmented" traffic has to be punted to LC CPU and that is subject to LPTS rate-limiting.

LU is doing that, more specifically reorder block is doing that, same
block which guarantees order of packets when single flow exceeds PPE
capacity. I think per flow fragmentation is shy of 2Mpps, but it'll
scale further by adding flows. At $job-1 we wanted to move some mobile
network related fragmentation out from software device into the edge
network MX, I was very dubious about the idea, but tested anyhow, and
was surprised to find out how fast it worked.

I'm sure there are some scenarios out there where fragmentation is
still unavoidable, or other solutions are even more expensive. I think
JNPR should be more vocal about this ability as market differentiator.
Cthulhu knows CSCO has been vocal to point out mcast benefits of ARK9k
over MX.





-- 
  ++ytti