[j-nsp] MX960 with 3 RE's?

[Saku Ytti]

On 15 January 2016 at 03:13, Christopher E. Brown
<chris.brown at acsalaska.net> wrote:
> When the same folks were asked about the 16XGE card and the 120G (and later 160G)
> performance it was indicated that there was an additional layer of logic/asics used to tie
> all 4 trios in the 16XGE to the bus and that these ASICs offloaded some of the bus related
> overhead handling from the TRIOs, freeing up enough capacity to allow each TRIO in the
> 16XGE to provide a full 40G duplex after jcell/etc overhead.


Sorry Christopher for being suspicious, but I think you must have done
some mistake in your testing.

Only difference that I can think of, on top of the multicast
replication, is that 16XGE does not have TCAM. But that does not
matter, as the TCAM isn't used for anything in MPC1/MPC2, it's just
sitting there.
MPC1/MPC2 can be bought without QX. You specifically metnion '3D-Q'.
If you were testing with QX enabled, then it's wholly different thing.
QX was never dimensioned to push all traffic in every port via QX,
it's very very much underdimensioned for this. If MQ can do maybe
~70Gbps memory BW, QX can't do anywhere near 40Gbps. So if you enable
QX or ingress+egress, you're gonna have very very limited performance.

I'm gonna need much more concrete data to believe 16XGE has higher
memory bandwidth or lookup performance than MPC2. If this information
is available to you, or if you can easily find it, I'd really really
appreciate it. This would be new information to me, and very much
interesting, it's just that right now, all data I have, is disagrees
with that.

-- 
  ++ytti