Re: Fwd: Re: WG Review: Transparent Interconnection of Lots of Links (trill)]

[Erik Nordmark <erik.nordmark at sun.com>]

[With reduced cc list]

Eric W Gray wrote:

>    The interesting thing here is that any changes in the encapsulation 
> of Ethernet
> frames would have drastic implications on all Ethernet equipment - 
> including NIC
> cards in Ethernet end stations. If this is not the case, then we are 
> talking about an
> encapsulation of a transport segment, rather than encapsulation 
> end-to-end. If that
> is what we are talking about, how does this "encapsulation" differ from 
> Martini?

Perhaps I can help answer this.

The encapsulation is not end-to-end; it is merely between the TRILL
devices. The draft-radia-rbridge draft shows the current thinking for
the encapsulation, with a header than in the case of carrying things
over Ethernet include
 - source rbridge MAC address (so that bridges between the rbridges
don't get confused)
 - next hop rbridge MAC address (to avoid proliferation should the
bridges flood the frame and multiple rbridges receive it)
 - ttl (to limit the damage from any temporary routing loops)
 - last hop rbridge (so that intermediate rbridges don't need to track
the location of each station)
 - Ethernet type

So this is quite different than the Martini approach.

Should TRILL later be expanded to apply to run over MPLS (something
which is not in the current charter) then I suspect we'd end up with an
encapsulation like Martini, and a link-state routing protocol carrying
the MAC addresses.

So the TRILL work is architecturally quite similar to the VPLS work in
L2VPN. The encapsulation that is safe and scalable for Ethernet is a key
difference.

   Erik