Re: [babel] Restarting nodes and seqno requests

[Toke Høiland-Jørgensen <[e-mail]>]

Juliusz Chroboczek <[e-mail]> writes:

>>>> My implementation of the randomisation only randomises the top 32 bits,
>>>> so there's still a stable identifier. Maybe it would make sense to
>>>> default to this behaviour?
>>> 
>>> That would break the optimisation of using the low bits of the IPv6
>>> address to encode the router-id.
>
>> Bird doesn't do that anyway.
>
> That's five lines of code, and saves almost half the traffic in pure
> IPv6 meshes.

Right, well the reason I didn't include it was that Bird already has a
notion of a router ID (from BGP), which is 32 bits and based on v4
addresses. And it would be breaking assumptions to just override that.

It could conceivably be a switch, though...

>>> I'm not sure how you'd avoid router-id collisions, though.
>
>> By chance? With 32 bits of randomness, even in a moderately sized
>> network the chance of collision is quite low (which means that it will
>> not happen until it is the most inconvenient, of course...) :)
>
> With 100 nodes, probability of collision is 10^-6.  With 1000 nodes, it's
> 10^-4.  With 2000 nodes, it's 5*10^-4.  With 5000 nodes, it's 3*10^-3.
>
> I think we're at the very edge of comfort, which is why I chose 64
> bits in the first place.

Well, the collision domain would not be all nodes, only those with the
same configured router ID (or v4 address absent any configuration).

But yeah, you're not wrong :)


BTW, in the case where there *is* a stable identifier (whether its
EUI-64 or a v4 address), randomising only parts of the router ID would
be a way to keep the stable identifier? For the v4/32 bit case,
randomising the 32 bits is the obvious solution; for EUI-64 you could
just randomise the FFFE part? That way you could gain the reboot support
but still have stable(-ish) identifiers so you could find your node?

-Toke

-Toke