Results for IPv8 Datasync
Simulation 1
Simulating 8, 48 and 168 hours of bartercast events (562, 5075, 15048 events accordingly) using different sync configurations
- We prove that the total number of syncs gives us the same convergence no matter how often we sync
- 1 sync per 4 hours is 58%
- 2 syncs per 8 hours is 59%
- A total of 15 syncs in 24 hours gives us a convergence of 97%
- To reach higher (98%, 99% or even 100%) we should waster a lot of upload and download bandwidth
- sent 201KiB per Node VS 524KiB, 532KiB, 1184KiB accordingly (260%, 264%, 589%)
- received 356KiB per Node VS 444KiB, 376KiB, 467KiB accordingly (124%, 105%, 131%)
- Huge gap between 6 syncs per 24 hours and 12 syncs per 24 hours (58% vs 95%)
- check with 8 syncs per 24 hours
Graphs
- Simulation for 200 nodes for 1 week (15k+ bartercast records)
- We need the system to scale to million of peers, so lets add semantic clustering. Simulate with groups of 50 peers (200 nodes in total, 1 week of bartercast records) (same number of syncs, different tactics)
- We got nice results from that, prove that we can keep local coverage the same while the total number of peers grows. Simulate with groups of 50 peers for 400 and 600 total peers
- OK but all simulations assumed that the peers are 100% online which is unrealistic. Go back to 200 nodes and simulate with 75% uptime...
|
