[c-nsp] Need for large buffers for 1-to-1 forwarding?

[Saku Ytti]

On (2012-11-22 01:05 +0100), Mathias Sundman wrote:

> As we only provide the customer with 1 client port on the remote
> device, there is only one ingress port that needs to forward to one
> egress port. Is there a need for larger buffers for such a usage or
> should any switch with enough pps forwarding capacity do the job for
> even the most demanding traffic?

You're golden. This is the optimum situation and easiest to handle in
hardware. There is no need for deep buffering.
Microbursts are issue when ingress is higher rate than egress or when there
are more ingress than egress interfaces.

> It's my believe that it is the client's own switch that aggregates
> multiple ingress ports that need large egress buffers on the
> interface connecting to our switch, right?

I'm losing you. If you have one port to customer, how can customer have
many ingress ports? Or were you first talking NET->CUST direction and now
CUST->NET direction? Surely the customer traffic is not balanced, that is
very untypical, almost always traffic is either inbound heavy or outbound
heavy.
But yes, if he customer has many ingress to single egress, it's customer
who is having issues.

> I'm also considering turning off mac-learning on the client's
> q-tunnel VLAN, as I can't why you would want to maintain a mac-table
> when you only have two ports to forward between, right? The client's
> switch should never be sending me anything unless he wants it to
> arrive at the remote site.

I don't see problem there. Just remember to be strict about allowed-vlan
stanzas in trunk ports.

-- 
  ++ytti