Carbon footprint and service coverage tradeoffs in geo-diverse sites

Abstract

Data centers are growing hulks on energy consumption, thus carbon emissions. It is critical to maintain the services, i.e., data intensive computing, while preserving the greenness in the data center management. However, the more data centers required data replicated in, the faster exponential increase in the services’ carbon footprint. What is worse, the carbon coefficient of the data center is time-dependent as the power source varies from local renewable farms to power grids, making the problem dynamic. To address the challenges above, in this paper, we consider the temporal and spatial characteristics of cross-region data replication and propose a data replication framework (Carbon Aware Geo-diverse DAta Replication) CAGDAR to reduce carbon footprint and maximize service coverage simultaneously. CAGDAR balances the carbon footprint of data centers, i.e., sites, while respecting storage capacity, taking into account the coverage of data copies and the locations. We evaluate CAGDAR with real carbon traces and show that CAGDAR can rationally allocate data replicas in the temporal-spatio domain, reducing the services’ carbon footprint by 26% and improving the service coverage by 33.5%.