This document describes simulations of peer-to-peer (P2P) networks using different protocols, configurations, and parameters. It discusses running many simulations to analyze the results using Tableau. Key aspects studied include average, maximum, and minimum aggregation functions, network size and topology, value distributions, and static versus dynamic network configurations. Real-world P2P networks tend to be either structured, unstructured, or hybrid depending on their topology and search capabilities.
6. When all nodes hold the
same value
Using Incremental Stats printed in average observer to see when
network is ‘settled’
control.avgo: 0 1.0 99.99999999999999 50 50.5 867.4354435651799 1 1
CDSimulator: cycle 0 done
control.avgo: 1 1.0 69.6938775510204 50 23.386122448979595 414.34298362076936 7 1
CDSimulator: cycle 1 done
control.avgo: 2 1.0 31.30612244897959 50 5.162040816326529 49.386130098853386 26 2
CDSimulator: cycle 2 done
control.avgo: 3 1.0 7.061224489795918 50 1.4040816326530612 1.6661425086486068 44 2
CDSimulator: cycle 3 done
control.avgo: 4 1.0 1.0 50 1.0 0.0 50 50
CDSimulator: cycle 4 done
Largest value
held by
any node
Cycle
Smallest value
held by
any node
Network
size
Mean
value
Variance
in values
Total nodes
holding
minimum
value
Total nodes
holding
maximum
value
Minimum function, linear distribution of values (1,100) , network size = 50
7. When all nodes hold the
same value
Using Incremental Stats printed in average observer to see when
network is ‘settled’
control.avgo: 0 1.0 99.99999999999999 50 50.5 867.4354435651799 1 1
CDSimulator: cycle 0 done
control.avgo: 1 1.0 69.6938775510204 50 23.386122448979595 414.34298362076936 7 1
CDSimulator: cycle 1 done
control.avgo: 2 1.0 31.30612244897959 50 5.162040816326529 49.386130098853386 26 2
CDSimulator: cycle 2 done
control.avgo: 3 1.0 7.061224489795918 50 1.4040816326530612 1.6661425086486068 44 2
CDSimulator: cycle 3 done
control.avgo: 4 50 1.0 0.0 50 50
CDSimulator: cycle 4 done
Largest value
held by
any node
Cycle
Smallest value
held by
any node
Network
size
Mean
value
Variance
in values
Total nodes
holding
minimum
value
Total nodes
holding
maximum
value
Minimum function, linear distribution of values (1,100) , network size = 50
1.0 1.0
8. When all nodes hold the
same value
Using Incremental Stats printed in average observer to see when
network is ‘settled’
control.avgo: 0 1.0 99.99999999999999 50 50.5 867.4354435651799 1 1
CDSimulator: cycle 0 done
control.avgo: 1 1.0 69.6938775510204 50 23.386122448979595 414.34298362076936 7 1
CDSimulator: cycle 1 done
control.avgo: 2 1.0 31.30612244897959 50 5.162040816326529 49.386130098853386 26 2
CDSimulator: cycle 2 done
control.avgo: 3 1.0 7.061224489795918 50 1.4040816326530612 1.6661425086486068 44 2
CDSimulator: cycle 3 done
control.avgo: 4 1.0 1.0 1.0 0.0
CDSimulator: cycle 4 done
Largest value
held by
any node
Cycle
Smallest value
held by
any node
Network
size
Mean
value
Variance
in values
Total nodes
holding
minimum
value
Total nodes
holding
maximum
value
Minimum function, linear distribution of values (1,100) , network size = 50
505050
9. Varying parameters
• Protocol {AverageFunction, MaximumFunction, MinumumFunction}
• Network size [2,50000]
• Distribution of values {Linear, Peak}, distribution parameters
• k, directed edges out of each node [1,8]
• Network Topography {WireKOut, WireINet, WireRegRootedTree, WireRingLattice,
WireStar, WireWS}
• Network dynamism {Static, Dynamic, Oscillating}, dynamic and oscillating parameters
21. Benefits of Simulation
Allows for a ‘sterile’ environment where approaches can be
designed to improve the components of a P2P network’s
functionality.
22. Benefits of Simulation
Test various approaches and topologies for their suitability in
the given scenario. This may imply a suitability for a real life
application.
41. Real World P2P
• Structured - Overlay has fixed topology, protocol
aims to ensure an efficient search of network for a
given resource.
ImageSRC:- “https://personalpages.manchester.ac.uk/staff/m.dodge/cybergeography/atlas/gnucleus_graph_large.gif”
42. Real World P2P
• Unstructured - No formal topology, easy to build.
Allow for localised optimisations and are highly
resilient against problems caused with high ‘churn’.
ImageSRC:- “http://courses.cse.tamu.edu/caverlee/csce438/hw/rand_graph.png"
43. Real World P2P
• Hybrid - Combination of P2P and client-server
models. Currently tend to have better all round
performance versus pure P2P and client-server.
ImageSRC:- “http://www.di.unipi.it/~hkholidy/projects/cids/CIDS%20in%20Pure%20P2P%20model.gif”
44. Real World P2P
• Spotify(2011) - Hybrid model, utilises centralised
servers, P2P network, and a local cache. Only non
web-based on-demand music streaming service.
ImageSRC:- “http://pansentient.com/2011/04/spotify-technology-some-stats-and-how-spotify-works/“
~8
~36~56
45. Real World P2P
• Spotify - Simple use case
ImageSRC:- “http://pansentient.com/2011/04/spotify-technology-some-stats-and-how-spotify-works/“
• User selects track
• If cached, play from there
• Else
• Request first 15secs from servers.
• Search P2P network for remainder of song.
46. Real World P2P
• Spotify’s P2P Network - Works like BitTorrent,
using server-side trackers and network queries to
locate peers. Returned peers are those that can
support playback.
Predictable listening habits - Most users listen to the
same music multiple times or listen to albums. Makes
predictive caching possible, reducing network traffic.
Editor's Notes
Development in a controlled environment
Allows for highlighting of functionality:
Means that development can be undertaken on the targeted functionality.
Testing specific functionality in a clean environment:
May indicate strengths/weaknesses.
May be applicable to real work scenarios.
OverSwarm - Taking inspiration from nature. Emulating natural networks, example: Pheromone trails.
DHTs - Using underlying data structure of a network to more efficiently support network queries under churn. More efficient than flooding network.