A hub location connected to 2 spoke sites through a L2
cloud with an Ethernet 2M CCT from carrier (this cloud does not support CoS). The
2 Spoke Sites have have 1M SDSL ccts ( from same provider).
We have business traffic to be prioritized over the non corporate
one. Main traffic flow is outgoing from the Hub and we are GRE Tunnels from hub
to Spokes (one per Spoke site) with HQoS applied. Concern here is value to
be configured on shaping (per each Tunnel) to proper considering the GRE + ETH
encapsulation overhead , avoiding to exceed the 2M of bandwidth
contractualized with the carrier (and consequently drops from their PE).
My belief is most Cisco shapers don't account for L2
overhead. So, you need to allow for that if the shaper doesn't.
Unfortunately, L2 overhead, as a percentage, varies per packet size. You
might shape for worst percentage or average percentage. What to allow for
depends on just how critical correct QoS operation is required. For
example, if you're dealing with VoIP packets, you'll probably want to allow
more toward the worse case percentage.
If you're DSL is using PPPoE, don't also forget to allow for the 8
bytes used by it.
Whether you need to also account for GRE overhead would, I believe,
depend on whether your shaper is configured on the tunnel interface or on the
physical interface. Like L2 overhead, GRE, as a percentage, would vary
per packet size. With GRE you also have the chance of fragmentation,
although if your device supports (and it's enabled) ip tcp adjust-mss should
preclude it (at least for TCP packets).
Citation - This blog post does not reflect original content from the author. Rather it summarizes content that are relevant to the topic from different sources in the web. The sources might include any online discussion boards, forums, websites and others.