MAS-DecStream uses LLM agents to cut edge latency by 97%

Researchers Sabeur Lajili and Zaki Brahmi have developed MAS-DecStream, a multi-agent scheduling framework for stream processing in mobile edge computing that leverages LLMs to dramatically improve resource allocation. The core innovation, called LLM-MR-CNP, extends the classical Contract Net Protocol (CNP) with semantic CFP (Call For Proposals) formulation, progressive context disclosure, multi-round proposal revision, and negotiation memory.

In experiments using the Alibaba ASI Trace, MAS-DecStream demonstrated impressive results: it reduced latency violations to just 3%, eliminated resource overcommitment entirely, achieved a 91% conflict-resolution rate across 20 agents, and improved utility by up to 22% compared to multi-round rule-based baselines. The framework's strength lies in its ability to handle qualitative runtime context through multi-round negotiation while maintaining deterministic constraints for hard resource and QoS requirements.