ART DeCO: Advanced Real-Time Decentralized Computation OffloadingMaster Thesis

24 June 2025, by Julius Lehmann

This work addresses the growing need for offloading processing tasks from low-power devices to spare compute capacity in a decentralized network. Existing methods, ranging from centralized FaaS platforms to Docker-based or proprietary-language systems, suffer from single‐point bottlenecks, high overhead, or limited accessibility. Decentralized scheduling schemes that use local decision making can ensure low latency and avoid single-point of failures. However, if no direct IP connections can be established, they struggle with NAT/firewall traversal and added latency from on-demand P2P connection establishment. We propose a decentralized scheduler that minimizes latency by efficiently creating suitable P2P connections in advance.

Supervised by:
Prof. Dr. Janick Edinger, Dr. Heiko Bornholdt, Anton Semjonov