Greenpeace’s Investigation of Amazon’s Rapidly Growing Cloud
by Gary Cook
May 11, 2015
Amazon has long kept Amazon Web Services' (AWS) operational details inside the proverbial black box, keeping details of its size and scale hidden even from its own investors until just this past month.
(Image above) Exterior view of a data center in northern Virginia.
Transparency has been a core component of Greenpeaces evaluation of data center operators since we began benchmarking performance in 2010, and we have seen significant improvement in energy transparency among data center operators over the past five years. Despite its market leading position among cloud providers, Amazon has long kept Amazon Web Services (AWS) operational details inside the proverbial black box, keeping details of its size and scale hidden even from its own investors until just this past month. While Amazon has finally provided more specific financial details on AWS, Amazon and subsidiary AWS unfortunately remain among the least transparent companies with regard to its environmental footprint, lagging far behind what Apple, Google, IBM, and nearly every other company in our 2015 evaluation provides.
For our past two evaluations of major internet company performance, Greenpeace has largely relied on third party analysis of AWSs server deployment as a basis for estimating the energy demand of AWSs data centers. However, this proxy data covered only a portion of AWS servers (EC2), and struggled to keep pace in measuring the rapid expansion of AWS infrastructure. Given the rapid growth we were seeing on the ground and the critical role that AWS occupies in building our online world, we filed a series of public information requests to shed new light on the scale of AWSs infrastructure and its rapidly increasing demand for electricity.
Emergency backup generators as a basis for estimating energy demand
All major data centers have multiple levels of backup power supply to ensure their operations remain uninterrupted in the event of fluctuations or disruption of the primary power supply. With a few recent exceptions, emergency diesel powered backup generators are deployed at all major data centers, typically at a scale that would generate enough power to provide a cushion (eg:20%) well above what is needed to fully power the data center for a significant period. Data center operators also deploy these generators with varying degrees of redundancy, typically at least one or two generators extra, to account for the failure of one or more generators (a backup to the backups).
While some data center operations, particularly colocation operators, will publish details on the number and size of backup generators installed, most companies do not. However, most local jurisdictions in the United States require permits be obtained for the installation and operation of diesel generators, given the particulate matter and other pollution that comes from operating them, even on an emergency basis, as they require regular testing to ensure they are in good working order. Greenpeace has used backup generator deployment as a basis for estimating data center power demand in the past, and continues to do so as needed if adequate reporting is not provided by the data center operator.
Investigation of AWS Energy Footprint
Greenpeace filed public information requests for emergency backup generator permits issued to Amazon or its relevant subsidiaries in each of the jurisdictions in the US where AWS currently has significant data center operations, which AWS designates as Availability Zones. For AWS Availability Zones we were able to secure information on emergency diesel generator permits (California, Ireland, Oregon, and Virginia) we have used these permits to form estimates of AWS data center power capacity. For Availability Zones based in Brazil, Japan, and Singapore, we continue to use third-party estimates of server deployment to estimate power capacity in each region. Data center operations associated with AWS CloudFront content distribution network have not been included in our estimates of AWS energy footprint.
Greenpeace Estimates of AWS Power Demand based on Backup Generator Permits
Its important to note that Greenpeaces estimates of data center power demand are in most cases based upon design capacity, not current consumption, and reflect announced facilities, which may not yet be fully operational. Estimating the potential power demand of a data center based upon emergency backup generators permits requires us to make several assumptions about their deployment:
- We assumed that generators are deployed with 1-2 redundant backups of the largest sized generator.
- We assumed that generators of a smaller size than the majority deployed are for water/fire suppression emergency backup.
- Estimates of data center power demand are based on an assumed 80% capacity of the primary set of deployed generators, based on size of the generator(s).
As expected, by far the largest of the AWS Availability Zones is US East, which is supported by data center infrastructure in Northern Virginia. Despite its significant existing size, US East also appears to be growing at a staggering rate. AWS has noted publicly that US East has more than 10 data centers. Based on the permits we reviewed, the total number of AWS data centers operating or under construction in Virginia currently stands at 23.
Most AWS data centers in US East are clustered together in mini-campuses of two to three data centers each, and we have grouped them accordingly in our report published today. This clustering of data centers appears to allow AWS to maximize its backup generator deployment, allowing it to deploy one set of redundant backup generators (the backups to the backups) to serve each data center cluster.
Despite previous estimates already placing well over half of the AWS servers in US East, and the absence of renewable electricity supply from the dominant utility Dominion Virginia Power, Amazon significantly expanded its data center infrastructure in Northern Virginia in 2014. Based on Greenpeaces analysis of emergency data center permits applied for in 2014 by Amazon subsidiary Vadata, AWS expanded the capacity of US East by over 200 MW, as much as the total amount of capacity in the rest of the US Availability Zones combined, bringing the total energy demand capacity of AWS facilities in Northern Virginia to 500 MW.
Greenpeace Analysis of AWS Availability Zones: Backup Generator Permits
|Location||AWS Availability Zone(s)||Permitting Entity||# of Data Centers||Greenpeace estimate of data center power capacity|
|Bay Area, CA||US West( N. CA)||A-100||2||36 MW|
|Dublin, Ireland||EU West||Amazon Data Services Ireland Ltd||3||65 MW|
|Northern Virginia||US East||Vadata||23||500 MW|
|Oregon||US West (Oregon) & GovCloud||Vadata||3||65 MW|
See the permitsbelowassociated with each of Amazon’s availability zones: