"MAS-NetCPO"

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:Offers a novel multi-timescale optimization framework for carbon-aware orchestration with online decision guarantees.

🏢実務担当者:Provides quantifiable carbon and cost reduction strategies for datacenter and network operations teams.

"MAS-NetCPO decomposes the coupled orchestration problem across three timescales. Network Agents perform fast routing and bandwidth coordination. Compute Agents make medium-timescale workload placement decisions. Energy Agents update carbon-aware power budgets at the slow timescale. The algorithm combines Lyapunov drift-plus-penalty optimization with asynchronous bandwidth consensus, which enables online decisions while maintaining bounded queue behavior under stochastic workload and carbon conditions.Simulations using workload, carbon, price, and WAN topology traces show that MAS-NetCPO improves the normalized benefit profile across latency, carbon emissions, electricity cost, and fairness. Compared with the network-only baseline, MAS-NetCPO reduces the latency index, carbon emissions, electricity cost, and sum of carbon emissions and electricity cost by 6.20\\%, 38.23\\%, 38.04\\%, and 38.19\\%, respectively. The $V$-sweep experiment illustrates the expected utility and queue tradeoff, the migration analysis shows interpretable carbon-aware spatial shifting, and the timing microbenchmark confirms that online decisions complete within the evaluated control intervals.Overall, MAS-NetCPO demonstrates that network-aware carbon orchestration can provide substantial sustainability benefits without ignoring latency and feasibility constraints. Future work will extend the framework toward prototype deployment, uncertainty-aware carbon forecasting, stronger privacy protection, larger multi-domain topologies, and lifecycle-aware carbon accounting."

• openalex https://doi.org/10.21227/cks7-bz59first seen 2026-06-21 05:01:56