Baseline numbers for Codex vs BitTorrent

[gmega]

Experiment

Parameters:

• Modes: BitTorrent, Codex Erasure-Coded, Codex Non-Erasure-Coded

• Swarm Size:

  • total size: 2, 8, 16, 32
  • seeders: 1, 2, 4, 8, 16
  • file size: $100$ MB, $1$ GB, $5$ GB

• Measurements:

  • completion distribution (track individual blocks?)

• Scenario:

  • start dedicated bootstrap node (Codex) or dedicated tracker (BitTorrent);
  • start all nodes in swarm;
  • upload files to seeders;
  • start the download at the leechers;
  • wait for download to complete at all nodes;
  • export event logs as CSV.

• We'll deem the performance of the system to be close to BitTorrent if we're able to:

  • we're no more than 20% slower than BitTorrent in all of the above scenarios.

[Zorlin]

I'd suggest something naughtier than Transmission - and more scriptable. Deluge and/or rTorrent would be my suggestion. Deluge is particularly aggressive over local networks, speaking from experience, so it may be the best choice overall given how easy it is to control over Python.

[gmega]

Hey @Zorlin, yeah, sounds good, and Deluge is libtorrent-based so that may also be a point in favour as we may aim for libtorrent compatibility in a not-so-distant future.

[mratsim]

Note that there are 2 libtorrent (rasterbar from arvidn and rakshasa)