[j-nsp] MX5-T VPLS fowarding problem

[Sergey Khalavchuk]

Hello,

On Fri, Mar 29, 2013 at 11:27 PM, Mathias Sundman <mathias at nilings.se> wrote:
> On 03/29/2013 12:40 PM, Caillin Bathern wrote:
>
> Can someone explain the benefits of using tunnel-services vs
> no-tunnel-services on the MX80 platform for VPLS services?

tunnel-services are just a hack to allow two lookups per one frame
received from mpls backbone using frame re-circulation:
1) label lookup to map frame to vpls instance,
2) mac lookup to forward frame to correct interface.

Recirculation means that every frame must be processed TWICE. And it
is possible that frame will cross fabric twice.
For this reason, when you enable tunnel-services for fpc, you must put
bandwidth limit (1g or 10g), and recirculated traffic will be policed.

On older fpc it is mandatory for VPLS.
On newer fpcs, it is possible to perform 1st lookup with IO manager on
ingress linecard, and avoid recirculation.

So, with vrf-table-label or with no-tunnel-services, there is single
lookup, no performance penalty, only profit.

Why "tunnel-services" may be needed?
1) older fpc (not mx).
2) SDH/ATM/local tunnels on ingress line-card doesn't allow IO manager
to perform 1st lookup, so tunnel services will be required.

> With a 10G backbone using 3-4 of the built-in 10G interfaces, and an
> expected VPLS use of 1-4G of bandwidth, what would be the recommended
> config?

use vrf-table-label of no-tunnel-services.

--
wbr,
Sergey Khalavchuk