165-simple-robust-voting.txt 5.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133
  1. Filename: 165-simple-robust-voting.txt
  2. Title: Easy migration for voting authority sets
  3. Author: Nick Mathewson
  4. Created: 2009-05-28
  5. Status: Open
  6. Overview:
  7. This proposal describes any easy-to-implement, easy-to-verify way to
  8. change the set of authorities without creating a "flag day" situation.
  9. Motivation:
  10. From proposal 134 ("More robust consensus voting with diverse
  11. authority sets") by Peter Palfrader:
  12. Right now there are about five authoritative directory servers
  13. in the Tor network, tho this number is expected to rise to about
  14. 15 eventually.
  15. Adding a new authority requires synchronized action from all
  16. operators of directory authorities so that at any time during the
  17. update at least half of all authorities are running and agree on
  18. who is an authority. The latter requirement is there so that the
  19. authorities can arrive at a common consensus: Each authority
  20. builds the consensus based on the votes from all authorities it
  21. recognizes, and so a different set of recognized authorities will
  22. lead to a different consensus document.
  23. In response to this problem, proposal 134 suggested that every
  24. candidate authority list in its vote whom it believes to be an
  25. authority. These A-says-B-is-an-authority relationships form a
  26. directed graph. Each authority then iteratively finds the largest
  27. clique in the graph and remove it, until they find one containing
  28. them. They vote with this clique.
  29. Proposal 134 had some problems:
  30. - It had a security problem in that M hostile authorities in a
  31. clique could effectively kick out M-1 honest authorities. This
  32. could enable a minority of the original authorities to take over.
  33. - It was too complex in its implications to analyze well: it took us
  34. over a year to realize that it was insecure.
  35. - It tried to solve a bigger problem: general fragmentation of
  36. authority trust. Really, all we wanted to have was the ability to
  37. add and remove authorities without forcing a flag day.
  38. Proposed protocol design:
  39. A "Voting Set" is a set of authorities. Each authority has a list of
  40. the voting sets it considers acceptable. These sets are chosen
  41. manually by the authority operators. They must always contain the
  42. authority itself. Each authority lists all of these voting sets in
  43. its votes.
  44. Authorities exchange votes with every other authority in any of their
  45. voting sets.
  46. When it is time to calculate a consensus, an authority votes with
  47. whichever voting set it lists that is listed by the most members of
  48. that set. In other words, given two sets S1 and S2 that an authority
  49. lists, that authority will prefer to vote with S1 over S2 whenever
  50. the number of other authorities in S1 that themselves list S1 is
  51. higher than the number of other authorities in S2 that themselves
  52. list S2.
  53. For example, suppose authority A recognizes two sets, "A B C D" and
  54. "A E F G H". Suppose that the first set is recognized by all of A,
  55. B, C, and D, whereas the second set is recognized only by A, E, and
  56. F. Because the first set is recognize by more of the authorities in
  57. it than the other one, A will vote with the first set.
  58. Ties are broken in favor of some arbitrary function of the identity
  59. keys of the authorities in the set.
  60. How to migrate authority sets:
  61. In steady state, each authority operator should list only the current
  62. actual voting set as accepted.
  63. When we want to add an authority, each authority operator configures
  64. his or her server to list two voting sets: one containing all the old
  65. authorities, and one containing the old authorities and the new
  66. authority too. Once all authorities are listing the new set of
  67. authorities, they will start voting with that set because of its
  68. size.
  69. What if one or two authority operators are slow to list the new set?
  70. Then the other operators can stop listing the old set once there are
  71. enough authorities listing the new set to make its voting successful.
  72. (Note that these authorities not listing the new set will still have
  73. their votes counted, since they themselves will be members of the new
  74. set. They will only fail to sign the consensus generated by the
  75. other authorities who are using the new set.)
  76. When we want to remove an authority, the operators list two voting
  77. sets: one containing all the authorities, and one omitting the
  78. authority we want to remove. Once enough authorities list the new
  79. set as acceptable, we start having authority operators stop listing
  80. the old set. Once there are more listing the new set than the old
  81. set, the new set will win.
  82. Data format changes:
  83. Add a new 'voting-set' line to the vote document format. Allow it to
  84. occur any number of times. Its format is:
  85. voting-set SP 'fingerprint' SP 'fingerprint' ... NL
  86. where each fingerprint is the hex fingerprint of an identity key of
  87. an authority. Sort fingerprints in ascending order.
  88. When the consensus method is at least 'X' (decide this when we
  89. implement the proposal), add this line to the consensus format as
  90. well, before the first dir-source line. [This information is not
  91. redundant with the dir-source sections in the consensus: If an
  92. authority is recognized but didn't vote, that authority will appear in
  93. the voting-set line but not in the dir-source sections.]
  94. We don't need to list other information about authorities in our
  95. vote.
  96. Migration issues:
  97. We should keep track somewhere of which Tor client versions
  98. recognized which authorities.
  99. Acknowledgments:
  100. The design came out of an IRC conversation with Peter Palfrader. He
  101. had the basic idea first.