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

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

Also, don't forget that unicast is possible for DHCPv6 (at the server's discretion) :).

Do we see any harm in using IPV6_USE_MIN_MTU?

If there is hop-by-hop communication (i.e., "direct route"), at least per the documentation I found the MTU of the interface is used (i.e., "direct route"). I guess if relay <-> server or relay <-> relay communication requires intermediate router hops, PMTU could apply as there are likely few end-points involved in the communication (just not sure how well PMTU works for UDP).

- Bernie

-----Original Message-----
From: [e-mail] [mailto:[e-mail]] On Behalf Of ????
Sent: Wednesday, May 07, 2014 12:49 PM
To: Ted Lemon
Cc: dhcwg; Ralph Droms; Bernie Volz (volz)
Subject: Re: [dhcwg] WGLC for draft-ietf-dhc-sedhcpv6-02 - Respond by May 18

At Wed, 7 May 2014 11:25:08 -0500,
Ted Lemon <[e-mail]> wrote:

> > One specific point: newer DHCPv6 implementations that support this 
> > protocol would have to use the IPV6_USE_MIN_MTU socket option (or 
> > something equivalent to it), for exact the same kind of reason why 
> > all modern DNS server implementations use it on UDP sockets.
>
> What reason is that?

To avoid packet drop (+ ICMPv6 too big, which would normally be ignored for UDP) at an intermediate router when you send a large DNS (or DHCPv6) message.  In case of DNS, DNSSEC made it more critical.
In case of DHCPv6, sedhcpv6 would make it more critical.  But, on thinking it in more detail, I realized I made the discussion too simplified (maybe you were trying to point it out?).  For DHCPv6 we'll need to discuss some more details:

- If the client only multicasts a message (to servers or relays), at
  least the client does not have this issue (but see the third bullet)
- So it's more about between servers and relays in practice.  In
  particular, since relays are not expected to be aware of sedhcpv6
  options, the implication will be less obvious for them and would be
  worth documenting explicitly.
- Even a client could unicast messages directly to a server after
  receiving the Server Unicast option, so it would still matter for
  clients in some rare cases.
- Regarding the similarity between DNS and DHCPv6, there's one
  important difference between them: in the case of DHCPv6 links (and
  MTUs) over which packets go would be less heterogeneous, so the risk
  of the drop due to this would be much smaller in practice (but that
  doesn't mean we can ignore it completely, so I think it's still
  worth discussing).

--
JINMEI, Tatuya