I withdraw my desire this proposal.  In Roster we wouldn't want these /16 network families---we just wanted to collapse some relays together when we reliably believe they have the same operator, and there's no reason to believe the majority of relays within a /16 are owned by the same person.

Ergo, Roster will forgo this kind of merging.

-V

On Friday, 5 February 2016, Karsten Loesing <karsten@torproject.org> wrote:
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

[Removing metrics-team@ to avoid cross posting.]

On 28/01/16 21:22, Tim Wilson-Brown - teor wrote:
>
>> On 29 Jan 2016, at 07:20, Roman Mamedov <rm@romanrm.net> wrote:
>>
>> On Fri, 29 Jan 2016 06:33:51 +1100 Tim Wilson-Brown - teor
>> <teor2345@gmail.com> wrote:
>>
>>> Tor already considers relays in the same IPv4 /16 to be in the
>>> same family.
>>
>> Maybe a step further in this would be to autoextend manually
>> declared families with all relays running on the same IPs of any
>> relays in the family. Dunno how complex or how useful this would
>> be. It could at least fix-up some outdated or missed
>> declarations.
>
> In Tor, or OnionOO?
>
> Tor already does this using the IP address whenever a path is
> built. If Tor added it on the relay side, then we'd bloat
> descriptors for no reason.

Agreed.

> If OnionOO added it, it would save OnionOO clients some work.

Let's consider this.  I'm pasting current definitions of related
Onionoo fields here, so that people can follow more easily:

 - "effective_family": Array of $-prefixed fingerprints of relays that
are in an effective, mutual family relationship with this relay. These
relays are part of this relay's family and they consider this relay to
be part of their family. Omitted if empty or if descriptor containing
this information cannot be found.

 - "alleged_family": Array of $-prefixed fingerprints of relays that
are not in an effective, mutual family relationship with this relay.
These relays are part of this relay's family but they don't consider
this relay to be part of their family. Omitted if empty or if
descriptor containing this information cannot be found.

 - "indirect_family": Array of $-prefixed fingerprints of relays that
are not in an effective, mutual family relationship with this relay
but that can be reached by following effective, mutual family
relationships starting at this relay. Omitted if empty or if
descriptor containing this information cannot be found.

Now, from reading this thread I can see us adding or extending the
following fields:

 - Extend "effective_family" to also include relays on the same IP
address or in the same /16.  I'd rather not want to do this, because
we wouldn't be able to say whether that other relay is in a mutually
declared family relationship or just runs on a nearby IP address.

 - Add new "network_family" field with fingerprints of all relays in
the same /16.  Plausible, but duplicates fingerprints that are already
in "effective_family".

 - Add new "network_family" field with only those fingerprints of
relays in the same /16 that are not contained in "effective_family".
"Tor considers these relays to be part of your relay's family, because
they have similar enough network addresses.  If you are running them,
please consider setting the family option."  Plausible, though not
trivial to grasp without further explanation.

 - Add new "extended_network_family" field with fingerprints of relays
in the same /16 as this relay or relays in "effective_family" and
"indirect_family", except for fingerprints in those two fields.  Also
plausible for the Roster use case to identify all relays close to the
family that the user may have omitted in their family definitions.
Not sure if this is necessary.

 - Add new "abandoned_family" field with fingerprints of relays that
declare this relay to be part of their family but that are not
contained in this relay's family declaration.  Looks like we never
considered this field before, but it might be useful to help relay
operators fix their family declarations.

Which of these fields would be useful to have?  "All of them" is not a
good response, because we shouldn't make Onionoo responses bigger if
nobody uses the new data.  But I'm happy to discuss use cases and then
add new fields as required.

All the best,
Karsten

-----BEGIN PGP SIGNATURE-----
Comment: GPGTools - http://gpgtools.org

iQEcBAEBAgAGBQJWtGubAAoJEJD5dJfVqbCrwDEIAMN/JCYq99J/H3AZKqkt3pLe
qvWP8uQxBfbnmxwOhOq4IFFCa1o+FpITOxmhZEuxVNGaqszBqSxFpDn62pjK8YCS
7Wi2IqUoZDIdHwLsJMgfrn+/HH4BoctTu0PzHWsZsmcdjJqPr8R+AP7WRZN3SM2W
/ML8AULWIwSUVmIfKD3iYM9RbFfxFeCARirDsAxC394z2ei06git4sJA5cSROx35
9IzqdpPyJoplYBRk7INCmr0bHNXvsIRODQ0n0QIJrIl1ESHpqhsy13fTo/1ndlKR
BUM2XCao0HABwpdBOrinfpybuGUSPXjrqw8expkUE+w2VuzOdkkNod1J3wgFyXc=
=KM0S
-----END PGP SIGNATURE-----
_______________________________________________
tor-relays mailing list
tor-relays@lists.torproject.org
https://lists.torproject.org/cgi-bin/mailman/listinfo/tor-relays