I inline a patch that specifies how voting should happen in proposal 236. The changes reflect a discussion I had yesterday with nickm during the Tor IRC meeting. BTW, while I like the simplicity of the new vote (just an integer), I'm afraid that this might hide any misconfigurations of the guardiness system in authorities. For example, maybe an authority is misconfigured and considers only 2 consensuses for guardiness data (instead of 6000 consensuses) and if it publishes only an integer percentage then we might not notice this misconfiguration. Maybe the number of consensuses parsed and the number of months considered should also be mentioned somewhere in the votes?
George Kadianakis <desnacked@riseup.net> writes:
I inline a patch that specifies how voting should happen in proposal 236.
The changes reflect a discussion I had yesterday with nickm during the Tor IRC meeting.
BTW, while I like the simplicity of the new vote (just an integer), I'm afraid that this might hide any misconfigurations of the guardiness system in authorities. For example, maybe an authority is misconfigured and considers only 2 consensuses for guardiness data (instead of 6000 consensuses) and if it publishes only an integer percentage then we might not notice this misconfiguration. Maybe the number of consensuses parsed and the number of months considered should also be mentioned somewhere in the votes?
I attach a new patch after discussion with Nick. Some improvements from the previous patch: - s/GuardAppearanceFraction/GuardFraction . However, if you can think of a better name for that keyword, I'm all ears. - Specify a new consensus method (hopefully I did it correctly) - Specify that we mean low-median when we say median.
On Sun, Aug 31, 2014 at 9:34 AM, George Kadianakis <desnacked@riseup.net> wrote:
George Kadianakis <desnacked@riseup.net> writes:
I inline a patch that specifies how voting should happen in proposal 236.
The changes reflect a discussion I had yesterday with nickm during the Tor IRC meeting.
BTW, while I like the simplicity of the new vote (just an integer), I'm afraid that this might hide any misconfigurations of the guardiness system in authorities. For example, maybe an authority is misconfigured and considers only 2 consensuses for guardiness data (instead of 6000 consensuses) and if it publishes only an integer percentage then we might not notice this misconfiguration. Maybe the number of consensuses parsed and the number of months considered should also be mentioned somewhere in the votes?
I attach a new patch after discussion with Nick.
Some improvements from the previous patch:
- s/GuardAppearanceFraction/GuardFraction . However, if you can think of a better name for that keyword, I'm all ears. - Specify a new consensus method (hopefully I did it correctly) - Specify that we mean low-median when we say median.
Merged it; thanks, George! -- Nick
participants (2)
-
George Kadianakis -
Nick Mathewson