A Distributed Framework for Carbon and Cost Aware Geographical Job Scheduling in a Hybrid Data Center Infrastructure
2016 IEEE International Conference on Autonomic Computing (ICAC) (2016)
July 17, 2016 to July 22, 2016
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/ICAC.2016.21
To mitigate the growing carbon footprints of data centers, many large IT organizations (e.g., Apple) have installed on-site renewables in their self-managed data centers. Meanwhile, to achieve low-cost global presence, these organizations often lease space and house their servers in geo-distributed colocation data centers, where they share the power (including renewables) with other tenants. Such sharing of renewable energy creates new challenges: how can an organization minimize its carbon footprint in colocations? While numerous studies have investigated geographic load balancing to minimize carbon emissions of data centers, these studies have primarily focused on self-managed data centers where all the renewables are solely dedicated to the data center operator. In this paper, we consider a practical hybrid data center infrastructure (including both self-managed and colocation data centers) and propose a novel resource management algorithm based on alternating direction method of multipliers, called CAGE (Carbon and cost Aware GEographical job scheduling) to reduce carbon footprints. CAGE dynamically distributes incoming workloads to geo-distributed data centers based on local renewable availability, carbon efficiency, electricity price, and also the energy usage of other tenants that share the colocation data centers. Our comprehensive simulation study and system experiment show the benefits of CAGE in terms of carbon footprint reduction: up to 36% compared to the state of the arts.
Servers, Carbon, Organizations, Carbon dioxide, Distributed databases, Renewable energy sources, Scheduling
A. H. Mahmud and S. S. Iyengar, "A Distributed Framework for Carbon and Cost Aware Geographical Job Scheduling in a Hybrid Data Center Infrastructure," 2016 IEEE International Conference on Autonomic Computing (ICAC), Wuerzburg, Germany, 2016, pp. 75-84.