Our cloud is thirsty!
|Power plant is "drinking" water
|240,000-gallon water storage tank at Google's data center in
Berkeley County, SC
|Google's data center in
The Dalles, OR, is "drinking" water at dusk
LA Times: California declares drought emergency
The Wall Street Journal: California governor declares drought emergency
“Businesses have been ordered to cut water use 35%” in certain areas (source)
Sacramento and Folsom have issued mandatory water usage restrictions (source)
Parts of 11 states, including Arkansas, California, Colorado, Hawaii, Idaho, Kansas, New Mexico, Nevada, Oklahoma, Texas and Utah, are designed as drought disaster areas (source)
Undoubtedly, data centers are extremely important for California…
As aforementioned, data centers consume water both directly and indirectly. Here, I would like to first draw the readers’ attention to the difference between water withdrawal and water consumption. The former refers to getting water from somewhere (e.g., public water facilities), whereas the latter refers to “losing” water (e.g., into the environment via evaporation) and producing waste water (e.g., into sewage systems). Both water withdrawal and consumption deserve our attention: water withdrawal causes an increasingly high pressure on the water supply side as the population continues to grow (e.g., thermal electricity generation accounts for 53% of fresh surface-water withdrawals in the U.S.), while water consumption threatens the long-term water sustainability and availability. In what follows, I will particularly focus on water consumption (also interchangeably referred to as water usage wherever applicable) which bears an immediate impact on the availability of groundwater.
Offiste/indirect water consumption: Indirect water consumption depends on electricity generation methods as well as cooling techniques and is mostly attributed to the evaporation process for steam condensation in cooling towers (typically required by nuclear and thermal electricity generation). While certain types of electricity (e.g., by solar photovoltaics and wind) consume virtually zero water, “water-free” electricity only makes up a very small portion of the total electric generation capacity (e.g., less than 10% in the U.S.). Overall, a non-negligible amount of water is “lost/consumed”: considering the power transmission loss but excluding hydroelectric, the U.S. national average water consumption is 1.8L/kWh (also referred to as Energy Water Intensity Factor or EWIF).
Combining both direct and indirect water consumption, data centers’ water footprints are now surfacing as a critical concern for future sustainability. Even with state-of-the-art facilities, Facebook's data center in Prineville, OR, is estimated to consume an average of more than 3.6L of water (both direct and indirect) per kWh of IT energy. The overall water footprint will be enormous considering the mega scale of data centers, and sometimes even exeed the capacity of local water utilities (as attested by the example of Microsoft's data center in Northlake, Illinois). To sum up, the growing trend of data centers’ water footprints can no longer be neglected and deserves careful attention from the research community.
Despite its emergence as a critical concern for data centers, water footprint has been largely and unfortunately neglected. Just as James Hamilton (Amazon) said in 2009, “water is tomorrow's big problem. The water consumption (in data centers) is super embarrassing. It just doesn't feel responsible. We need designs that stop using water.” In recent years, some large IT companies such as Microsoft and Google have made an impressive step towards reducing their water footprints. To summarize, the existing approaches to saving water in data centers can be classified as follows.
Leveraging outside cold air: Data centers built in cold regions (e.g., Dublin) can leverage ‘‘free’’ cooling by pushing outside cold air into data center computer rooms where hot air and cool air mix to remove heat generated by high-density servers. Thus, cooling towers, where water evaporates to reject the heat into the environment, can be eliminated. In some cooling systems (e.g., employed by Facebook's data center in Prineville, Oregon), outside air will mix with water sprayed by misting nozzles before entering computer rooms to keep appropriate operational temperature and humidity. Thus, cooling systems combining outside air with evaporative cooling still consume a non-negligible amount of water: as of March, 2013, the cooling water usage at Facebook's data center in Oregon still reaches 0.52L per kWh of IT energy.
Using non-potable water: Google and Microsoft have been using recycled/waste/sea water in lieu of potable water for cooling their servers. The water will be treated prior to entering their data centers’ cooling system.
Reusing warm water for heating: Warm water returned from data centers may be reused for heating offices and redidential buildings. Thus, cooling towers are not necessarily required.
These engineering-based approaches, however, mostly concentrate on improving cooling facilities and suffer from one or more of the following limitations. First, they require appropriate climate conditions and/or desirable locations that are not applicable for all data centers (e.g., “free air cooling” is ideally suitable in cold areas such as Dublin where Google has one data center). Second, they do not address indirect off-site water consumption. Last but not least, some of these approaches, such as building water treatment facilities, often require substantial capital investments that may not be affordable for all data center operators.
Media attention: Besides the existing engineering efforts made by large data center operators, data centers’ water footprints have received much media attention. I'll list a few representative media reports as follows, and a more comprehensive list can be found here: list of media reports.
Recognizing that water efficiency is emerging as a growing priority for data centers but the pace of its innovation is lagging far behind its energy counterpart, the objective of my research is to reduce the water footprints of data centers via software-based online resource management, which is complementary to the existing data center research and also fundamentally differs from engineering-based water-saving techniques (e.g., improving cooling facilities done by Google and Microsoft). I have done some preliminary work to optimize data centers’ water efficiency. In what follows, I'll list some of my recent work.
Optimizing Water Efficiency in Distributed Data Centers:
I begin my research by investigating
the characteristics of data centers’ water consumption. In particular, I
identify temporal and spatial diversities of
data center water usage effectiveness (WUE): data centers’
WUE changes over time and also over location.
The temporal diversity can be explained by noting that temporal changes in outside environment (e.g., temperature/humidity) will affect the usage of cooling water and that power plant's electricity production consists of time-varying mixes of energy fuel sources (each type of which requires different amount of water for a unit electricity generation, e.g., thermal electricity consumes a large amount of water while wind electricity consumes virtually zero water). Readers may refer to Facebook's dashboard at https://fbpuewue.com to view temporal variations of direct WUE (although Facebook is not using cooling towers). The figure to the left shows a snapshot of time-varying energy fuel mixes in California, which will thus lead to a time-varying indirect WUE (also referred to Electricity Water Intensity Factor, which measures the water consumption per unit of electricity production).
The spatial diversity can be easily understood: power plants in different places use different energy fuel mixes and/or cooling towers, resulting in spatial differences in EWIF (or indirect WUE); data centers in different places have different temperatures/humidities/cooling techniques/server configurations, etc., all of which will jointly affect direct WUE.
By exploiting the temporal and spatial diversites of data centers’ water efficiency, I proposed a new geographic load balancing algorithm (GLB) to dynamically schedule workloads to water-efficient data centers while satisfying a set of constraints such as cost and delay. To my best knowledge, the new resource management solutions (despite in its infancy) represent the first research efforts to address the emerging issue of data centers’ water footprints.
S. Ren, “Optimizing Water Efficiency in Distributed Data Centers,” International Conference on Cloud and Green Computing (CGC, acceptance ratio: 19/80=24%), 2013. [PDF]
M. A. Islam, K. Ahmed, S. Ren, and G. Quan, “Making Data Center Less ‘Thirsty' via Online Batch Job Scheduling,”
Tech. Report, 2013. [PDF]
Here, I'll maintain a list of research publications related to (or not so related to) data centers’ water footprints. I'll try to keep this list as complete as possible.
(excluding engineering-based methods done by large data center operators, e.g., improving cooling facilities)
R. Sharma, A. Shah, C. Bash, T. Christian, and C. Patel. Water efficiency management in datacenters: Metrics and methodology. In ISSST, 2009.
E. Frachtenberg. Holistic datacenter design in the open compute project. Computer, 45(7):83-85, July 2012.
C. Bash, T. Cader, Y. Chen, D. Gmach, R. Kaufman, D. Milojicic, A. Shah, and P. Sharma. Cloud sustainability dashboard, dynamically assessing sustainability of data centers and clouds. HP Labs Tech. Report (HPL-2011-148).
D. Alger, Grow a Greener Data Center, Cisco Press (ISBN-13: 978-1587058134), 2009.
If you have any questions and/or comments, please let me know and I appreciate your valuable feedback that'll help me tremendously improve my work.
I'm actively seeking external collaboration and funding supports. If you're interested in helping save water for our cloud, please contact me at email@example.com.
* Updated on January-15-2013