Michael S. Keller via Dng <dng@???> wrote:

>> As the router supports PD (Prefix Delegation), this means it must be running a DHCPv6 server, and it looks like your client is using this instead of autoconfig.
>> Can you use something like Wireshark (or it’s command line sibling, tshark) to capture RA packets and decode what flags it is setting for each prefix advertised ?
>
> For the IPv6 setup in my router, it offers these Assigned Types:
>
> ND proxy
> SLAAC+Stateless DHCP
> DHCPv6
> SLAAC+RDNSS

That’s not a bad selection of options - so many just offer “on” and “off” !

> It was set to DHCPv6 for months.
>
> I just changed it to SLAAC+Stateless DHCP, and left the client machine set to "auto" in /etc/network/interfaces.
>
> It can now get an IPv6 address this way, though it looks very different from what it received before when it hadh "dhcp" in the interfaces file and with the Assigned Type set to the prior DHCPv6.

Out of curiosity, in what way does it look “very different” ?
You should get an address (or addresses*) in the same prefix, though of course some other information will change to reflect the different assignment method in use. As I said, I’m no guru with this, so it’s of interest to see how “stuff I haven’t done” looks (I’ve only used SLAAC and static configs).

* Don’t forget, IPv6 supports any interface having an arbitrary number of addresses, in an arbitrary number of prefixes. As a minimum, every interface will have a link-local address, and to work with anything but it’s immediate neighbours must also have at least one more address.

> So far dd+nc over IPv6 doesn't stall, even after fifteen minutes of continuous sending.
>
> I'm not ready to declare it solved - I want to give it a few days.

That’s a result then. But it doesn’t explain why you were getting the pauses.
Configuration via DHCP should give pretty well the same experience (at least in this static environment) as SLAAC. So I can’t really think of any reason for it. I could see a “faulty” DHCP implementation causing addressing problems, but that would relate to the lease lifetimes, and most definitely not cause any form of “consistently n seconds after a transfer starts” problem.



>> * Mostly because a certain vendor of a mobile OS is so determined that
>> no-one will use DHCP on a mobile (and yes, there are valid arguments for
>> this) that they not only don’t include a DHCPv6 client in the OS, but
>> also coerce the hardware manufacturers to filter DHCPv6 packets at the
>> hardware layer in the mobile chipsets to make sure no third party can
>> fill the gap. The fact that there are genuine reasons, including legal
>> requirements, for many to use DHCPv6 is apparently not a good enough
>> reason to not block it entirely for everyone to prevent “misuse”.
>> As a result, if you run a managed network with DHCP and no autoconfig,
>> then a large chunk of devices will not work on it.
>
> And now my Pixel gets an IPv6 address.

Two results for the price of one :-)


Regards, Simon