Abstract

Centralized Learning Management Systems (LMS) struggle with scalability, bandwidth bottlenecks, and single points of failure during peak usage periods like examinations or global synchronous lectures. This paper proposes a novel hybrid Peer-to-Peer (P2P) architecture specifically designed for collaborative e-learning environments, leveraging institutional Super-Peers for administrative control, content validation, and DHT index maintenance alongside distributed student Standard-Peers for resilient resource storage and delivery. The framework integrates three key innovations: (1) Kademlia-based Distributed Hash Tables (DHT) for logarithmic resource discovery of chunked educational content (videos, datasets, documents); (2) an adapted tit-for-tat incentive mechanism with reputation scoring to mitigate free-riding; and (3) Conflict-free Replicated Data Types (CRDTs) enabling real-time collaborative tools like shared whiteboards and document editing without central coordination. Evaluation used PeerSim simulations across 30 runs (100-500 nodes, 500×500m² campus area, 20% hourly churn), validated against NS-3 wireless models and filtered Kaggle network traffic patterns representing lecture streaming. Results demonstrate superior performance: 76% average latency reduction (115±12 ms vs 480±25 ms client-server baseline, F=245.3, p<0.001), 7.3× aggregate throughput improvement (88±8 Mbps vs 12±2.1 Mbps, F=312.7, p<0.001), and 96±2% resource success rate under high churn (vs 62±4.2%, F=189.4, p<0.001). Energy efficiency improved 60% (180 mJ vs 450 mJ per task). Ablation studies confirm critical contributions: removing incentives caused 40% throughput degradation; replacing DHT with broadcast search spiked latency to 350±45 ms (3× increase). Raw data, simulation code, and statistical analyses available at github.com/user/p2p-elearning.