1 Filename: 160-bandwidth-offset.txt
2 Title: Authorities vote for bandwidth offsets in consensus
3 Author: Roger Dingledine
10 As part of proposal 141, we moved the bandwidth value for each relay
11 into the consensus. Now clients can know how they should load balance
12 even before they've fetched the corresponding relay descriptors.
14 Putting the bandwidth in the consensus also lets the directory
15 authorities choose more accurate numbers to advertise, if we come up
16 with a better algorithm for deciding weightings.
18 Our original plan was to teach directory authorities how to measure
19 bandwidth themselves; then every authority would vote for the bandwidth
20 it prefers, and we'd take the median of votes as usual.
22 The problem comes when we have 7 authorities, and only a few of them
23 have smarter bandwidth allocation algorithms. So long as the majority
24 of them are voting for the number in the relay descriptor, the minority
25 that have better numbers will be ignored.
29 One fix would be to demand that every authority also run the
30 new bandwidth measurement algorithms: in that case, part of the
31 responsibility of being an authority operator is that you need to run
32 this code too. But in practice we can't really require all current
33 authority operators to do that; and if we want to expand the set of
34 authority operators even further, it will become even more impractical.
35 Also, bandwidth testing adds load to the network, so we don't really
36 want to require that the number of concurrent bandwidth tests match
37 the number of authorities we have.
39 The better fix is to allow certain authorities to specify that they are
40 voting on bandwidth measurements: more accurate bandwidth values that
41 have actually been evaluated. In this way, authorities can vote on
42 the median measured value if sufficient measured votes exist for a router,
43 and otherwise fall back to the median value taken from the published router
46 3. Security implications
48 If only some authorities choose to vote on an offset, then a majority of
49 those voting authorities can arbitrarily change the bandwidth weighting
50 for the relay. At the extreme, if there's only one offset-voting
51 authority, then that authority can dictate which relays clients will
54 This problem isn't entirely new: we already have the worry wrt
55 the subset of authorities that vote for BadExit.
57 To make it not so bad, we should deploy at least three offset-voting
60 Also, authorities that know how to vote for offsets should vote for
61 an offset of zero for new nodes, rather than choosing not to vote on
62 any offset in those cases.
66 First, we need a new consensus method to support this new calculation.
68 Now v3 votes can have an additional value on the "w" line:
69 "w Bandwidth=X Measured=" INT.
71 Once we're using the new consensus method, the new way to compute the
72 Bandwidth weight is by checking if there are at least 3 "Measured"
73 votes. If so, the median of these is taken. Otherwise, the median
74 of the "Bandwidth=" values are taken, as described in Proposal 141.
76 Then the actual consensus looks just the same as it did before,
77 so clients never have to know that this additional calculation is
82 The Measured values will be read from a file provided by the scanners
83 described in proposal 161. Files with a timestamp older than 3 days
86 The file will be read in from dirserv_generate_networkstatus_vote_obj()
87 in a location specified by a new config option "V3MeasuredBandwidths".
88 A helper function will be called to populate new 'measured' and
89 'has_measured' fields of the routerstatus_t 'routerstatuses' list with
90 values read from this file.
92 An additional for_vote flag will be passed to
93 routerstatus_format_entry() from format_networkstatus_vote(), which will
94 indicate that the "Measured=" string should be appended to the "w Bandwith="
95 line with the measured value in the struct.
97 routerstatus_parse_entry_from_string() will be modified to parse the
98 "Measured=" lines into routerstatus_t struct fields.
100 Finally, networkstatus_compute_consensus() will set rs_out.bandwidth
101 to the median of the measured values if there are more than 3, otherwise
102 it will use the bandwidth value median as normal.