Re: [trill] Thoughts on active-active edge

[Thomas Narten <[e-mail]>]

"Tissa Senevirathne (tsenevir)" <[e-mail]> writes:

> > Proposed rule: R1, R2, and R3 should use the pseudonode nickname
> >  when encapsulating unicast, but use their own nickname when
> >  encapsulating multidestination.
>  
> > Result: R8 will see endnode changes for E, but not very frequent; R8
> >  will only see changes when traffic from E alternates between
> >  multidestination and unicast.
 
> [Answer] This depends on the traffic Radia. You cannot assume all
>  multicast hash to Node R1 and all unicast to R2 and R3 and hence
>  multicast unicast address ping pong is rare. In the contrary, you
>  could have heavy multicast stream and a unicast stream both hashing
>  in to the same RBridge RB1.

I'm with Tissa on this one. I'm wary about making assumptions about
"likely traffic patterns" and then making architectural decisions
based on that. We could easily get burned down the road.
 
> > Question: Are perceived endnode moves a problem?  Why?
 
> [Answer] Yes it is big time.  Unlike ancient devices modern devices
>  to hardware based learning. When address association start moving,
>  it will constantly bombard the CPU with new address or address move
>  notifications.

One thing that would help I think would be to actually list the issues
associated "MAC flapping" (i.e., where the FDB entries for a given
destination Etherent MAC addr keeps flip flopping) in terms of best
egress RB.

Are the issues exactly the same as for regular bridges? Or are there
additional nuances? What breaks or causes pain when entries flap?

Possible issues:

1) the paths packets take for a given destination could switch between
2 (or more) paths. This could lead to out-of-order packet
delivery. TCP doesn't like this. OAM may not like it either. (It's one
thing to have occasional router flaps, its another to have them be a
steady-state feature.)

2) Some implementations require software "help" to update FDB
entries. I.e., changes are not done 100% by the ASICs. While that may
change in the future, this could be an issue today for some
implemntations. How common is this problem? Do we have a sense?

What else? 

Thomas