[tor-bugs] #7009 [Tor]: Handle unstable relays better

[Tor Bug Tracker & Wiki]

#7009: Handle unstable relays better
---------------------------------+------------------------------------------
 Reporter:  arma                 |          Owner:                  
     Type:  project              |         Status:  new             
 Priority:  normal               |      Milestone:  Tor: unspecified
Component:  Tor                  |        Version:                  
 Keywords:  tor-relay, SponsorJ  |         Parent:                  
   Points:                       |   Actualpoints:                  
---------------------------------+------------------------------------------

Comment(by nickm):

 Okay, here's the third batch.

 Since we don't appear to have microdescriptor consensus archives, I faked
 some. I took all the votes from the month in question, and built a table
 mapping descriptor digest to voted-on v8 microdescriptor digest. (I used
 v8 because everything has a v8; this will throw off results a little
 though.) And then I used another script to postprocess the regular
 conensuses into microdescriptor-consensus-like things, by removing the
 descriptor digest from the "r" line and adding a corresponding "m" line.
 Of course, the signatures aren't valid on those, but that shouldn't make
 the measurements wrong.

 This table includes a set of "condiff2" measurements. These used an
 algorithm similar to "conndiff", except that they try harder to minimize
 "r" and "s" lines. The s lines now include only the changed flags
 (prefixed with a + or -), and the r lines now include only the changed
 fields (with the r suffixed with the field indices). This is still a
 kludge, even bigger than the last kludge.

 {{{
 EXPERIMENT 3: ERSATZ MICRODESCRIPTOR CONSENSUSES
 gz: mean 326449. median 326641
 bz2: mean 277180. median 277835
 xz: mean 297899. median 298140
 uncompressed: mean 920397. median 921975
 diff_gz: lag 1: mean 50356. median 49460
 diff_gz: lag 2: mean 81220. median 80293
 diff_gz: lag 4: mean 129828. median 128991
 diff_gz: lag 6: mean 171407. median 170568
 diff_gz: lag 8: mean 206859. median 206379
 diff_bz2: lag 1: mean 47274. median 46564
 diff_bz2: lag 2: mean 75232. median 74429
 diff_bz2: lag 4: mean 119625. median 118888
 diff_bz2: lag 6: mean 157949. median 157200
 diff_bz2: lag 8: mean 190432. median 189854
 diff_u_gz: lag 1: mean 169543. median 166231
 diff_u_gz: lag 2: mean 232762. median 229722
 diff_u_gz: lag 4: mean 301298. median 298904
 diff_u_gz: lag 6: mean 343474. median 341967
 diff_u_gz: lag 8: mean 371618. median 370834
 diff_u_bz2: lag 1: mean 150566. median 147908
 diff_u_bz2: lag 2: mean 205454. median 202992
 diff_u_bz2: lag 4: mean 266339. median 264134
 diff_u_bz2: lag 6: mean 307525. median 306157
 diff_u_bz2: lag 8: mean 335207. median 334631
 diff_e_gz: lag 1: mean 29085. median 28736
 diff_e_gz: lag 2: mean 47697. median 47365
 diff_e_gz: lag 4: mean 77821. median 77752
 diff_e_gz: lag 6: mean 104169. median 104026
 diff_e_gz: lag 8: mean 126932. median 126427
 diff_e_bz2: lag 1: mean 25963. median 25667
 diff_e_bz2: lag 2: mean 41830. median 41676
 diff_e_bz2: lag 4: mean 67435. median 67218
 diff_e_bz2: lag 6: mean 89650. median 89523
 diff_e_bz2: lag 8: mean 108744. median 108210
 condiff_gz: lag 1: mean 43394. median 32943
 condiff_gz: lag 2: mean 69116. median 56412
 condiff_gz: lag 4: mean 104882. median 112103
 condiff_gz: lag 6: mean 131878. median 140190
 condiff_gz: lag 8: mean 152803. median 159291
 condiff_bz2: lag 1: mean 38842. median 29785
 condiff_bz2: lag 2: mean 60713. median 49797
 condiff_bz2: lag 4: mean 90748. median 97287
 condiff_bz2: lag 6: mean 113359. median 120583
 condiff_bz2: lag 8: mean 130878. median 136453
 condiff2_gz: lag 1: mean 34492. median 23456
 condiff2_gz: lag 2: mean 52918. median 38968
 condiff2_gz: lag 4: mean 75948. median 87076
 condiff2_gz: lag 6: mean 91545. median 104040
 condiff2_gz: lag 8: mean 102485. median 112830
 condiff2_bz2: lag 1: mean 31025. median 21394
 condiff2_bz2: lag 2: mean 46673. median 34578
 condiff2_bz2: lag 4: mean 65934. median 76072
 condiff2_bz2: lag 6: mean 78876. median 89747
 condiff2_bz2: lag 8: mean 87823. median 96913
 }}}

 Next I'll do this one last time, with microdescriptor consensuses *and*
 weight rounding. Then conclusions.

-- 
Ticket URL: <https://trac.torproject.org/projects/tor/ticket/7009#comment:5>
Tor Bug Tracker & Wiki <https://trac.torproject.org/>
The Tor Project: anonymity online