[rbridge] header rough consensus

[J. R. Rivers]

I wanted to point out three addition things...

First, since one of TRILL's goals is L2 multi-pathing, I think we'll
find that information from layers other than L2 will be used for
distributing frames (ie. IP addresses, TCP/UDP ports, FC_IDs) as is done
for L2 port channels and IP multipathing today.  This means that
implementations will likely have to parse past the TRILL header in all
cases.

Second, care must be taken surrounding the expectations of the extension
header.  Since it's usage is undefined, there is a good chance that
implementations will evaluate/forward frames with extensions in the slow
path.  I would vote that extensions get evaluated/defined in their own
groups and get real Ethertypes.  Particularly scary is that the
extensibility in the proposal is currently 124B which is roughly twice
the size of a min sized Ethernet frame.

Lastly, TRILL frames traversing legacy L2 networks would likely travel
down a port bundle.  Since TRILL is a new Ethertype, most existing
bundle distribution systems will use a the L2 DA/SA/VLAN to select links
(they won't be able to get to higher level protocols).  We might want to
allow an RBridge to have multiple synonymous MAC addresses that can be
used as the Ethernet SA to facilitate load balancing.  The transmitting
RBridge could select a source address based on a load balancing
algorithm.

JR



> -----Original Message-----
> From: rbridge-bounces at postel.org 
> [mailto:rbridge-bounces at postel.org] On Behalf Of Silvano Gai
> Sent: Sunday, May 06, 2007 9:26 AM
> To: rbridge at postel.org
> Subject: [rbridge] header rough consensus
> 
> 
> Trying to summarize the results, my understanding is:
> 
> Solution 1
> - Silvano
> - Dino
> - Joe
> - Eric
> 
> Solution 2
> - Donald
> 
> Not sure about:
> - Dinesh
> - Radia
> 
> Am I correct?
> Did I miss anybody?
> 
> -- Silvano
> 
> _______________________________________________
> rbridge mailing list
> rbridge at postel.org
> http://mailman.postel.org/mailman/listinfo/rbridge
>