The RiPiT project is researching, building, and deploying information
services that digest the complex dynamics of power grids to enable
computing to reduce its carbon emissions impact.
The power grid is a complex, increasingly dynamic large-scale
infrastructure with temporal and spatial variation in carbon-emission
exceeding 10-times. As more renewables are deployed and grids are decarbonized, these ratios are increasing.
Computing's extraordinary agility (seconds, even milliseconds) provides a
unique ability to respond and adapt, or plan and shift logistically to reduce
carbon emissions. But to do so intelligently and productively,
computing needs dynamic information that reflects the impact of power consumption.
Computing also has a responsibility for climate change, as the cloud, high speed
networking, and a growing universe of edge and client devices drive a rapidly
growing consumption of power and carbon footprint.
Plan:
RiPiT will provide a succession of information services and APIs
that enable a variety of computing devices, systems, users to intelligently plan and execute
compute load shifting in time and place, data and storage management,
and battery charging. Such intelligent planning requires real-time and forecast
data, but RiPiT faces several challenges to meet these needs.
- Most power grids do not provide sufficient and timely information, so
RiPiT is exploring a range of analytic and inference techniques to provide
guidance to computing systems. We expect this information to improve in accuracy, spatial and temporal
resolution, and timeliness as the project progresses.
- Carbon emissions depend on the detailed temporal and
geographic pattern of consumption, combined with the realized grid mix and
dispatch. RiPiT will provide attribution services that allow consumption
to be translated into carbon emission estimates.
- Efficient interaction with applications, and providing useful information in the
presence of uncertainty is a difficult challenge. If you are interested in working with us to
develop productive APIs, please contact us.
The RiPiT Metrics System includes several parts:
- Part I: Examples of Grid
Data and Metrics reported by real power grids (MISO, ERCOT,
SWPP, CAISO) -- with varying definition, coverage, latency, and
resolution. This highlights the challenge in "making sense" of
data available in the power grid.
- Part II: RiPiT-designed Metrics and Summaries that provide
insights in thinking about how to reduce carbon emissions for
computing. Grid behavior is distilled relative to workload properties
(flexibility, temporal alignment, etc.) to suggest both where you
might be able to achieve the largest reductions, and the likely
difficulty of doing so. This provides a digested perspective on the
complex dynamics of power grids. These new metrics are are computed
from models, prediction, and grid provided data. They guide initial
steps to find the most promising opportunities to reduce carbon
emissions impact.
- Part III: A RiPiT API that allows intelligent loads to express flexibility, evaluate potential carbon-emissions impact of load adaptation, register adaptation, and calculate impact. This is designed for applications to use the detailed workload structure and create intelligent management to reap reductions in carbon emissions, and document those reductions.
For background, see the : Zero Carbon Cloud (ZCCloud) Project which has demonstrated that
- Cloud computing is growing rapidly and consumes nearly 10 percent of grid power in several areas.
- To reduce its Scope 2 carbon emissions, one viable strategy is to
adapt computing load spatially and temporally. Shifting should target when there are more renewables available, and also shift to increase the grids ability to absorb renewable generation.
- Shifting of datacenters loads of 10s to 100s of megawatts can be done effectively, eliminating carbon
emissions and creating zero-carbon computing.
RiPiT is supported in part by a generous grant from VMWare Research and the National Science Foundation.
RiPiT is part of the Zero-carbon cloud (ZCCloud) Project and the Large-scale Sustainable Systems Group (LSSG) at the University of Chicago .
People:
Tristan Sharma, Varsha Rao, and Liuzixuan (Peter) Lin,
Andrew A. Chien (UChicago),
Jiaqi Chen, Joseph Gorka,
Line Roald,
(U Wisconsin), and
Rich Wolski
(UCSB),
Former: Cameron Fiske, Fan Yang, Jeremy Archer (UChicago)
|