New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Thesis: Using distributed blockchains to cultivate trust in a peer group #2533
Comments
"Cultivating online trust in complex self-organising systems" Nice title/storyline. Background story: we tried multichain, needs churn resilience, Tribler future-proof architecture means creating a generic "trust cultivation" module. Essential requirement is that this can be re-used in other systems, proven to work in Tribler. Self-reinforcing trust is a more cold and rational term. Cultivating is more associated with organic and human touch. Interesting. |
Design: why create records every 10min, not every 10MByte ? Experiments brainstorm:
October goal: generate scenario, based on real behavior from our crawl. Either: give existing records or do 1 download/second for 3 hours == 10k downloads of Multichain records. Perhaps with @qstokkink close the self-reinforcing loop and select non-freeriding nodes for tunnels. |
Public ongoing thesis text : https://www.sharelatex.com/project/573db2896f2d756841e3def8 |
|
ToDo for thesis roadmap:
|
Experiments:
Start to define thesis committee |
Discovery experiment (Gumby) scenario:
|
Trace-driven simulator, inside Jenkins: https://jenkins.tribler.org/job/Tribler/job/Experiments/job/multichain/job/walker/ws/ Conclude: 1 chapter on simulator. |
To clarify: the exploration is actually monotonically increasing, the bug is in the plot script, which renders this in a weird way. Should still be fixed ofcourse. |
Planning: Dec - finish simulator plot. then document in thesis. Chapter 2: please add the onion routing context as key use-case. There the 1:13 contribution ratio for default Tor-like setup is an unsolved problem. We aim to solve it for the somewhat easier 1:5 ratio scenario. "Unconditional Random walk vs reputation-directed walk". Danger: how is this work different? Repeating the existing Dimitra experiments? "3.2.2 Statefull vs stateless walk" incorrect terminology. Either walk to Dispersy neighbor in overlay or to a Multichain neighbor in overlay+shared datastructure. Dispersy chains the contacting of neighbors into a random walk, with probability of .5% for a teleport to a trusted bootstrap server. Revisiting the non-backwards compatible design with deep walker integration. Unified primitive for:
Thus together forming trusted peer discovery suitable for NAT constrained Internet connections. Each live edge request simply includes the records the sender already knows about. The notation can be a simple (start block number, last block number). The first live-reponse message always returns the last record, with the most recent transaction. This ensures that each subsequent re-visit we can reply with either fresh records, historical records, or an empty record. |
Recent changes: design chapter "3.2 Record discovery" Simulator is focussed on block discovery; without churn, no block creation, and infinite local storage. Please make the load-balancing plot more dramatic: some nodes are going to crash. For instance, X-axes has all 700+ nodes sorted by their message-load, Y-axes shows their requests handled. Each node in the system is depicted as a datapoint, with a dramatic curve for the first dozen nodes on the left-hand side (no log-log scale needed). Implementing "live edge walker" in Gumby estimated to be roughly 1 week. {Gumby darknet mode broken, possibly use clearnet; simple seeding records} |
Good. Finish "Live edges" work in Dispersy, repeat nice load balancing graphs. |
msc_thesis___Pim_Veldhuisen (1).pdf
Finish most thesis chapters or finish all experimental results? ToDo johan reminder: committee form |
ToDo johan reminder: committee form Theory + https://www.google.nl/search?q=the+shadow+of+the+future Branching attack, mention double spending. 4.2 Experiments; Would call it emulation, not simulation. why continuous discovery from numerous sources ? attack-resilience... 5.2.1 Teleport probabilities. Still difficult to reproduce. underspecified. how many nodes with this 10k blocks? step time? how many connected neighbors? all their blocks when visiting somebody Figure 5.3: The block collection process using different teleport probabilities. |
Page 3, try to say in first line why Tribler is relevent here. We do not yet know workload of a true rewarding cooperation mechanism. Explore basic scalability and cost functions. Explore: storage cost of records for 10k to 1Million blocks? |
msc_thesis___Pim_Veldhuisen (5).pdf Chapter 6: how do we conclude this is really correct? verified? Can a link with cooperation be made? |
Leveraging blockchains to establish cooperation - Pim Veldhuisen - Master's thesis.pdf Final thesis from delft repository |
φ is dependent upon how important reputation is. If the only purpose of reputation is to encourage the shadow of the future, maybe 0.05 is ok. But depending on the ratio of attackers/buggy clients versus honest nodes, the value of reputation may be different to different network participants. |
Thesis: Using distributed blockchains to cultivate trust in a peer group
The text was updated successfully, but these errors were encountered: