Multi-Agent Scheduling with LLM-Assisted Contract Net Negotiation for Stream Processing in Mobile Edge Computing
06:00 · August 15, 2026 · arXiv cs.AI RSS
Stream-processing systems increasingly operate across heterogeneous mobile edge--cloud infrastructures, where workload volatility, resource contention, and stringent quality-of-service (QoS) requirements complicate decentralized scheduling. This paper proposes \emph{MAS-DecStream}, whose main contribution is \emph{LLM-MR-CNP}: an extension of the classical Contract Net Protocol with semantic CFP formulation, progressive context disclosure, multi-round proposal revision, negotiation memory, and deterministic validation. Edge-cluster agents refine natural-language offloading proposals from local observations, predicted resource states, and qualitative runtime context, while hard resource and QoS constraints remain deterministic. Experiments derived from the Alibaba ASI Trace evaluate the extension at three levels: single- versus multi-round CNP, rule-based versus LLM-assisted refinement, and fixed-model single- versus multi-round negotiation. Under the evaluated configurations, MAS-DecStream reduces latency violations to 3\%, eliminates resource overcommitment, reaches a conflict-resolution rate of 0.91 with 20 agents, and improves utility by up to 22\% over the multi-round rule-based baseline. A separate 25-case evaluation shows model- and prompt-dependent accuracy--cost trade-offs. The results provide initial evidence that multi-round CNP refinement is the principal protocol-level gain, with LLM assistance adding value for qualitative and uncertain runtime context.
Summary
Stream-processing applications for domains such as smart cities and industrial IoT now run across heterogeneous mobile edge–cloud infrastructures, where rapid workload changes, partial system visibility, and strict quality-of-service constraints make centralized scheduling impractical. Independent local decisions frequently produce resource contention and latency violations that cannot be foreseen from any single cluster’s observations. MAS-DecStream addresses this setting through a decentralized multi-agent architecture in which each edge-cluster agent negotiates task offloading with its neighbors.
The core contribution is LLM-MR-CNP, an extension of the classical Contract Net Protocol. It augments the protocol with semantic call-for-proposals generation, progressive disclosure of workload and QoS context, multi-round proposal revision supported by negotiation memory, and a hybrid validation layer that keeps hard resource and deadline constraints under deterministic control while allowing large language models to interpret qualitative runtime information. Agents therefore refine natural-language offers across rounds before a final allocation is committed, rather than committing after a single announcement–award cycle.
Evaluation on traces derived from the Alibaba ASI dataset compares single-round versus multi-round negotiation, rule-based versus LLM-assisted refinement, and different model–prompt combinations. Under the tested conditions the approach lowers latency violations to 3 percent, removes resource overcommitment entirely, achieves a conflict-resolution rate of 0.91 among 20 agents, and improves overall utility by as much as 22 percent relative to a multi-round rule-based baseline. Separate experiments indicate that multi-round refinement accounts for most of the protocol-level gain, while LLM assistance contributes additional value when qualitative or uncertain context must be interpreted. The authors release the implementation, prompts, and evaluation data to support replication.
Why it matters
High technical depth and novelty in LLM-assisted multi-agent coordination directly applicable to Dutch edge-computing and IoT research; supports ethical, decentralized AI deployment relevant to EU contexts and SME adoption.
