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

[Saku Ytti]

On 15 January 2016 at 01:39, Christopher E. Brown
<chris.brown at acsalaska.net> wrote:
> The 30Gbit nominal (actual 31.7 or greater) limit per trio applies to the MPC1 and 2 cards
> but the quad trio interconnect in the 16XGE is wired up diff with additional helpers and
> can do the full 40G per bank.

There is actually one difference in MPC1/MPC2 and 16XGE. MPC1/MPC2 do
binary multicast replication, while 16XGE does unary.

So if your box is full of 16X10GE, multicast stream hops from NPU to
NPU, so every NPU gets the stream replicated at different time (i know
stock exchanges running MX and masturbating over nanoseconds in
switching latency, while they are doing three orders of magnitude
larger jitters in MX replication).
While if box is full of MPC1/MPC2, each card or NPU, will send copy to
two other NPU, for dramatically more efficient replication. (Of course
the theoretically correct way is to replicate in fabric, but I'm not
paying one cent for the right solution for this feature which should
be just checkbox item at best)

This also means that if you run multicast, trio needs 80Gbps of fabric
capacity for those 40Gbps of WAN capacity. DPCE/ichip has this, DPCE
is 40Gbps, but has 80Gbps of fabric capacity.
SCBE and MPC2 has this, 80Gbps of WAN capacity and 160Gbps of fabric capacity.
While 16XGE needs just same fabric capacity as WAN capacity.

-- 
  ++ytti