[j-nsp] JUNIPER POLICER and CoS Shaping Rate

[Saku Ytti]

On (2012-10-03 23:54 -0300), GIULIANO (WZTECH) wrote:
 
> set class-of-service interfaces ge-0/0/1 unit 530 shaping-rate 20m
> 
> 
> The output traffic rates 19.2~ Mbps only (using MRTG and SNMP
> statistics and graphics).

What do you see on fully congested 100M ethernet, no policers/shapers on
MRTG/SNMP? Hint, you don't see 100M.

Why? Because we can't count.

I've not yet seen system which shows actual wire use in graph, which is
only thing we network engineers care about.

If your line is 20M and you're sending 1500B (best case). You'll see
19.51Mbps L2 speed (what you are calculating right now) or 18.99 L3 speed.

That is without VLAN header. Add one VLAN header and those numbers drop to
19.46Mbps L2 and 18.94Mbps L3.
QinQ, MPLS etc, it'll all decrease speed (L2) you're measuring.


I know that Juniper MX shaper does the right thing and calculates L1 rate.
As does say Cisco ES20. But Cisco ES+ shaper calculates L2 speed, which is
extremely bad default, luckily it is configurable.

Why is shaper measuring L2 speed bad? 
Router - Distribution_L2 -100M- Customer

If you shape router VLAN to 100M and then prioritize VoIP on it, if router
is calculating L2 speed, you'll end up dropping VoIP packets in the
physical pipe, which certainly is calculating L1 speed.


If you use 'traffic-control-profile', you can manually set overhead
accounting, which you need if far-end overhead is different than local
interface.
-- 
  ++ytti