[tor-talk] Location-aware persistent guards

[Robert Ransom]

On 10/15/12, Maxim Kammerer <mk at dee.su> wrote:
> On Mon, Oct 15, 2012 at 7:58 PM, Robert Ransom <rransom.8774 at gmail.com>
> wrote:
>> https://bugs.torproject.org/2653
>
> Thanks, I must have read this ticket some a year or so ago — explains
> the identical sorting approach (forgot about weighting). Strangely,
> though, I don't see an option to add myself to CC.

Mike Perry took the TICKET_MODIFY permission away from ‘normal’ Trac
users because one moron misinterpreted a change to one ticket's
keywords (https://trac.torproject.org/projects/tor/ticket/7078#comment:9).
 (I don't see how that change will prevent normal users from *reading*
changes to ticket keywords, but whatever.)

>> The hardest part is load-balancing among the possible entry guards.
>
> A simple extension along the lines of what you wrote in the first
> comment would work fine, no? E.g., Sort a list by H(seed | nodeID |
> <1...w_nodeID>), where w_nodeID is the corresponding weight,
> approximated to an integer using some granularity setting, and take
> the top 3 distinct nodeIDs. With persistence, you can also ignore
> nodeIDs newer than the seed's timestamp, although not doing that is
> not critical.

My suggestion would be quite inefficient if entry guards' weights vary
widely.  Sorting a long list is even more inefficient, and would need
to be done in data-independent time if (as you suggest) the sort keys
are secret.


Robert Ransom