Presentation Information
[B-7-27]Carbon-Aware Job Scheduling Under Network Bandwidth Constraints
〇Waka Teramoto1, Shugo Nakamura1, Takashi Sano1, Hirotada Honda1 (1. Toyo Univ)
Keywords:
Carbon-Aware Computing,Binary Integer Programming
With the rapid proliferation of renewable energy, the carbon intensity (CI) of power grids has come to vary significantly across regions and time periods. As data center power consumption continues to rise, Carbon-Aware Computing — which leverages these spatiotemporal variations in CI to control computational job execution — has attracted growing attention. Existing studies have primarily focused on optimizing execution site selection or job start times independently, and integrated optimization incorporating network routing has not been sufficiently explored. To address this gap, we formulate a novel Carbon-Aware job scheduling problem that simultaneously optimizes execution site, communication path, and start time under network bandwidth constraints, with the goal of minimizing total CO2 emissions. We introduce a binary placement variable y(k,i,p,τ) indicating whether job k is assigned to site i, path p, and time slot τ. CO2 emissions are formulated using cumulative carbon intensity at each time slot and power consumption at each site and link. The problem is cast as a 0-1 integer programming problem minimizing total system-wide CO2 emissions, subject to bandwidth, site capacity, and deadline constraints.