[j-nsp] Cut-through forwarding statistics on QFX5100

[Saku Ytti]

On 23 June 2017 at 08:39, Vincent Bernat <bernat at luffy.cx> wrote:

Hey,

> Related question: why isn't cut-through the default mode? Is there any
> downside of this mode (apart from forwarding invalid frames which
> shouldn't matter much nowaday)?

You'll pass errored frames to the network, as you'll only know about
errors after you've sent the frame. In reality this isn't such a big
deal, as we don't guarantee non-brokenness of frames anyhow, generally
devices, even switches, regenerate ethernet frame (because they have
to, if VLAN stack, COS etc changes, so they do it always for
simplicity) on egress and  calculate new FCS on it, so if there is
broken memory on device, we calculate correct FCS on the frame we just
broke.

But also upside on cut-through is very limited, you obviously only can
do cut-through if egress happens to be completely empty, even in
modestly loaded box, you'd often operate store-and-forward. So you
have to have very specific application to see consistent cut-through
behaviour.

-- 
  ++ytti