Re: [dhcwg] dhcpv6-24: Elapsed Time option

> The DHC WG Meeting Minutes that Ralph published stated that the WG did 
> want to have load balancing for relays. I specifically asked about this 
> issue and I was promoting that we don't do it. What is your recollection?

I don't remember, but that doesn't sound incorrect.  (I think that's as 
wishy-washy an answer as I can give, but I'm open to suggestions... :')

> If we extend load balancing to relays, and load balancing includes the 
> ability to use the "elapsed-time", aren't we really asking for trouble?

Maybe I'm making this too complicated.   If the relay agent always 
_expands_ the set of servers to which it relays the packet as the retry 
time increases, it's safe.   It's only unsafe in the case where the set of 
servers (A) to which the relay agent relays at time=t is _disjoint_ with 
the set of servers (B) to which the relay agent relays at time=t+1.   As 
long as set A is a subset of set B, we're fine.   This is probably 
something that we can specify pretty safely.

> I would like to keep relays very simple in DHCPv6 and make them simply 
> forward the packets. All of the work should be done by servers, IMHO.

I think this is a good motivation, and I don't mind going this route.   
Here's what I suggest: we specify that the time in the retry time field in 
a Request message that immediately follows an Advertise message should be 
the time since the first Solicit was sent, not the time since the first 
Request message was sent.   This covers our patooties in the event that we 
change our mind on the Relay agent issue later.

_______________________________________________
dhcwg mailing list
dhcwg@ietf.org
https://www1.ietf.org/mailman/listinfo/dhcwg