Rational and effective computational workload management for carbon-emissions requires insight into the complex dynamics of power grids. These dynamics are driven by fluctuating generation (e.g. renewable, driven by weather, cloud, time of day) and load (time of day/week, season, events) that combine to produce major shifts in carbon-emissions. To manage workloads according to their carbon emission potential requires a new generation of metrics capturing and modeling the relationship between grid carbon emissions and cloud workload flexibility (e.g. long-running VMs, user-facing services, mission-critical services, persistent services) and/or migration costs (e.g. vm, data gravity, etc.)
In the UCSB RACELab we are developing the software and machine learning technologies necessary to collect, process, and distribute (at scale) carbon emissions metrics. The goal of the project is to allow datacenter providers (as well as other large-scale power consumers) to get measurements and projections of the carbon footprint associated with region electricity usage in near real time.
The UCSB RiPiT project releases its main software research artifacts as open-source software under a permissive license and does not intend to file patents on them.
The project is leveraging the UCSB Aristotle Cloud and the Eucalyptus open-source cloud infrastructure. In addition, the project will be using the Experimental Data Center located in the UCSB Institute for Energy Efficiency for empirical evaluations of the results it generates.