Proposal 167: Vote on network parameters in consensus

[Roger Dingledine]

Filename: 167-params-in-consensus.txt
Title: Vote on network parameters in consensus
Author: Roger Dingledine
Created: 18-Aug-2009
Status: Open
Target: 0.2.2

0. History


1. Overview

  Several of our new performance plans involve guessing how to tune
  clients and relays, yet we won't be able to learn whether we guessed
  the right tuning parameters until many people have upgraded. Instead,
  we should have directory authorities vote on the parameters, and teach
  Tors to read the currently recommended values out of the consensus.

2. Design

  V3 votes should include a new "params" line after the known-flags
  line. It contains key=value pairs, where value is an integer.

  Consensus documents that are generated with a sufficiently new consensus
  method (7?) then include a params line that includes every key listed
  in any vote, and the median value for that key (in case of ties,
  we use the median closer to zero).

2.1. Planned keys.

  The first planned parameter is "circwindow=101", which is the initial
  circuit packaging window that clients and relays should use. Putting
  it in the consensus will let us perform experiments with different
  values once enough Tors have upgraded -- see proposal 168.

  Later parameters might include a weighting for how much to favor quiet
  circuits over loud circuits in our round-robin algorithm; a weighting
  for how much to prioritize relays over clients if we use an incentive
  scheme like the gold-star design; and what fraction of circuits we
  should throw out from proposal 151.

2.2. What about non-integers?

  I'm not sure how we would do median on non-integer values. Further,
  I don't have any non-integer values in mind yet. So I say we cross
  that bridge when we get to it.