Published August 25, 2026 | Version v1.0.0

raxonros/marl-swarm: MARL-Swarm v1.0.0

Description

Source code, trained policies, evaluation harness and analysis scripts for the paper
"Same Quality, Cheaper and Fairer: Multi-Agent Reinforcement Learning for Joint Bitrate
and Origin Selection in Hybrid CDN-P2P Swarms".

Includes the event-driven CDN-P2P simulator, the Phase 4 platform of six node processes
serving real video, the B1-B5 baselines, IPPO/MAPPO training code, 141 checkpoints, the
frozen evaluation protocol, 209 tests, and swarm-lab, a packaged interface that estimates
origin-traffic savings on a user-supplied scenario.

Every number in the paper is regenerated from a committed artefact by a script.

Files

raxonros/marl-swarm-v1.0.0.zip

Files (27.4 MB)

Name Size
md5:6077593ac9fe6574a48371e45dce462c
27.4 MB Preview Download

Additional details

Related works