Rewarding good behavior

[synctext]

Determine the contributions of peers using our blockchain.

Track freeriders using MultiChain, compatible with anonymous seeding, relaying, and downloading mode.

Goal: superior performance compared to tit-for-tat, reduce freeriding and promote cooperation.
usage: accounting of bytes for seeding and Proxy bandwidth donations
Security: Sybil attack hardened, misreporting, lying, cheating, white washing and spoofing-proof.

Previously: Vlad thesis project, now Ewout!

[synctext]

Ideas for experiments (and thesis sections):

• usual intro, problem description (reward goodness), design, experiments (open Q: libswift&Dispersy background)
• 1 TByte seeding with Libswift (fuse, performance and item scalability)
  operational code, no Jenkins experiment yet.
• Libswift Top-N proportional sharing experiment with 1TByte of seeding
  http://jenkins.tribler.org/jenkins/job/libswift-priority/7/ operational code
  • AIM: demonstrate the reward mechanism
• BarterCast3 dissemination of Libswift (1 TByte) seeding statistics (double signed record creation only)
  http://jenkins.tribler.org/jenkins/job/Experiment_Swift_Effort_Link/ws/dispersy_standalone.py/*view*/ Initial code
• BarterCast3 and Libswift proportional sharing experiment
  • AIM: demonstrate dissemination linked to reward mechanism
  • initial BarterCast3 record creation, regular update mechanism as bytes increase, Libswift connection close/broken, then disseminate (after X MByte?, after X minutes?)
  • Cycle-based effort/reliability tracking is currently dictating the BarterRecord creation, should primary be byte-accounting now
  • two leechers compete for bandwidth
  • never met seeder before (strangers, indirect reciprocity only)
  • show the performance reward for seeding
• 16-ish Tribler instances on DAS4 ?
  • AIM: "building up trust in virtual identities without direct contact"
  • scalability: use Boudewijn Top250?
  • local reputation calculations?
  • re-use Dimitra biased reputation-based edge traversal?
• Converted opportunistic freerider experiment
  • AIM: cooperation is rewarded almost instantly
  • three classes of peers (die-hard freerider, opportunistic freerider, contributor)
  • die-hard freeriders never uploads and get low or zero performance (starvation mechanism?)
  • conversion mechanism of "opportunistic freerider", which only uploads if forced to
  • behavioral change of "opportunistic freerider" into a contributor (improving low to high performance)
• (abandoned idea, it's too complex) Append-only signatures for superior security and 100% accuracy
  • AIM: total fairness through completely accurate accounting
  • prior contributions instead of partial info
  • for partial info you know exactly the missing info fraction and how out-of-date you are; using the lastest BarterCast timestamp

[synctext]

5 directories with experiment code by Vlad: https://github.com/vladum/tribler-experiments
The 1TByte seeding code: https://github.com/vladum/lfs-libswift
System test: http://jenkins.tribler.org/job/Experiment_Libswift_No_Reciprocity/22/

[synctext]

Lessons learned after this msc thesis:
proportional bandwidth sharing is difficult and complex.
It works, but it should be avoided if possible.
Thus a form of access-control with a "request denied, insufficient credits/reputation" + "ongoing proxy help cancelled, a higher reputation peer pre-emptied you".

[VoR0220]

Hi synctext. I'm interested in learning more about what you learned from doing proportional bandwidth sharing. Where can I look to learn? Thanks in advance.

[synctext]

@VoR0220 The final thesis report and overview can be found here:
https://github.com/Tribler/tribler/wiki/BarterCast3

But: the key lesson is don't try this (not in report). Better approach is to deny service if a user is below "freerider level" and just let all the non-freerider sharing according to network conditions.

[VoR0220]

Thanks a ton @synctext I can't wait to read what you found and learn from you :)

[synctext]

Incentives and Prosocial Behavior good work with 1800 cites

[synctext]

Closing this ticket and work moves to "strongest possible reputation system" #31
plus #2571