Re: problem statement [was Re: New Version Notification for draft-hinden-ipv4flag-00.txt]

[Lorenzo Colitti <[e-mail]>]

On Sun, Nov 19, 2017 at 11:25 PM, Nick Hilliard <[e-mail]> wrote:

> > Let's not forget about battery life. According to the numbers in RFC
> > 7772, on a phone even just sending DHCPv4 packets every 2 minutes could
> > reduce battery life by ~8% compared to a completely idle device.
>
> is there any guarantee in this situation that any other apps would not
> be chattering in the background over ipv6 anyway?


No, but a lot of that can be dropped in firmware more cheaply. Waking up to
send a packet is more expensive.

Also, the whole internet isn't a mobile phone network, and what might be
> relevant or appropriate on a cell network might not be in any way
> relevant to another type of network.


The most relevant of which being, of course, DHCPv4 itself, which is not
used on cell networks. :-)


> One relatively straightforward way of dealing with extraneous dhcpv4
> packets would be to create a new dhcpv4 reply option hinting to the
> requester to cease DHCPv4 requests on the interface in question, or to
> slow down the request rate from one every two minutes to one every X
> minutes where X is network defined.
>

That would certainly work. Like any other DHCP option it would require
constant polling on the part of the client, but if X is high enough that
doesn't matter. Also, unlike the RA option, it makes it impossible to
change the flag with a network push, but in the long term that doesn't
matter either.

A harder problem is how to add this to DHCPv4 in a way that doesn't
negatively affect 20+ years of deployed client implementations. It can't
just be another option in the OFFER or ACK packets, since those contain the
client's IP address. I don't think it can be a new option in the NAK since
RFC 2131 says that if a client gets a NAK it should restart, and I bet
there are clients out there that restart immediately without waiting (which
would cause a loop). A new message type, with consequent modifications to
the DHCPv4 state machine? That seems pretty heavyweight.