ARTFEED — Contemporary Art Intelligence

LLM-Assisted Contract Net Protocol for Stream Processing in Mobile Edge Computing

ai-technology · 2026-08-15

A recent study published on arXiv (2608.12371) introduces MAS-DecStream, a framework designed for multi-agent scheduling in stream processing within mobile edge computing. The key innovation is LLM-MR-CNP, an enhanced version of the Contract Net Protocol (CNP) that incorporates large language models (LLMs) for creating semantic call-for-proposals (CFPs), facilitating progressive context sharing, enabling multi-round proposal adjustments, maintaining negotiation memory, and ensuring deterministic validation. This framework tackles issues in diverse mobile edge-cloud setups, such as workload fluctuations, resource competition, and strict quality-of-service (QoS) standards. Experiments utilizing the Alibaba ASI Trace assess the extension across three dimensions: single versus multi-round CNP, rule-based versus LLM-supported refinement, and fixed-model single versus multi-round negotiation. The paper falls under AI and distributed systems, indicating its potential for improving resource management in edge computing scenarios.

Key facts

  • Paper arXiv:2608.12371 proposes MAS-DecStream for stream processing scheduling.
  • Main contribution is LLM-MR-CNP, an extension of Contract Net Protocol.
  • LLM-MR-CNP includes semantic CFP formulation, progressive context disclosure, multi-round proposal revision, negotiation memory, and deterministic validation.
  • Targets heterogeneous mobile edge-cloud infrastructures with workload volatility and resource contention.
  • Edge-cluster agents refine natural-language offloading proposals from local observations and predicted resource states.
  • Hard resource and QoS constraints remain deterministic.
  • Experiments use Alibaba ASI Trace.
  • Evaluation covers single- vs multi-round CNP, rule-based vs LLM-assisted refinement, and fixed-model single- vs multi-round negotiation.

Entities

Institutions

  • arXiv
  • Alibaba

Sources