Re: [dhcwg] WGLC for draft-ietf-dhc-sedhcpv6-02 - Respond by May 18

["Bernie Volz (volz)" <[e-mail]>]

It is interesting that Jinmei raised this issue given the following presentation he gave:

http://www.v6.wide.ad.jp/Presentations/ietf49-udpapp.pdf

"- But ... At least every BSD kernel always performs PMTU discovery.
 - not necessary if we can always rely on the kernel’s behavior."

I do agree that the DNS problem is different because (a) there may not be many attempts per 'transaction' so working out the PMTU can be difficult and (b) there could be LOTS of end-points and that may exceed the kernel's cache by the time a request comes around again.

So, I guess best to just rely on the Kernel PMTU support.

Getting back to this draft, I think pointing out that this can result in large packets and that for some communication (relay <-> relay/server, client/server unicast), PMTU support is critical. And, that where packets are larger than the v6 min mtu (1280), firewalls in the DHCPv6 communication path that drop fragments must not be used.

We can consider your request in 3315bis - I created http://trac.tools.ietf.org/group/dhcpv6bis/ticket/92.

- Bernie

-----Original Message-----
From: Ted Lemon [mailto:[e-mail]] 
Sent: Wednesday, May 07, 2014 3:42 PM
To: Bernie Volz (volz)
Cc: dhcwg; 神明達哉; Ralph Droms
Subject: Re: [dhcwg] WGLC for draft-ietf-dhc-sedhcpv6-02 - Respond by May 18

There are two ways to do PMTU discovery.   One is to just send the packet, and if we get an ICMP Packet Too Big packet back, we set the path MTU to the reported size and try again.   The other is to detect it in the application layer protocol.   In this case, that would be DHCP, probably the DHCP relay agent and server.   This would be necessary in the case of a firewall that drops ICMP Packet Too Big messages.

I think it would make sense to add a recommendation in 3315bis that operators configure their firewalls to pass ICMP Packet Too Big messages.   I don't think we should go any further than that.