Google Apps: Energy Efficiency in the Cloud

Table of Contents Executive Summary

Executive Summary..........................................1 We know that cloud-based computing can reduce IT capital costs, reduce labor
Background: the cloud-based
costs, and enhance productivity. And a growing body of evidence shows the
services that we analyzed................................1
Why is the cloud more energy efficient?........2
cloud is also remarkably efficient.1 Last year we released a paper on the energy
How much can you reduce energy use savings from using Gmail instead of locally hosted email.2 Here, we expand on
and pollution by migrating to the cloud?.......3 that analysis to include the cloud-based office applications of Google Apps. Our
A real-world case study: GSA...........................3 estimates, which we’ve supported with a case study from a Google Apps client,
Conclusions........................................................4 show that migrating basic IT applications to Google Apps significantly reduces
Appendix: Notes on assumptions energy consumption and carbon emissions.
and methodology..............................................5
Based on our analysis, a typical company or organization that migrates to the
cloud could:
• save an estimated 68–87% in energy for its office computing
• reduce similar amounts of carbon emissions
These findings are consistent with a case study presented in this paper of the
actual savings achieved by the U.S. General Services Administration (GSA),
a Google Apps client with approximately 17,000 users. By switching to
Google Apps, GSA reduced server energy consumption by nearly 90% and
carbon emissions by 85%. We estimate the cost savings from this reduction in
energy use will be $285,000 annually (a 93% reduction).

Background: the cloud-based services that we

analyzed
While cloud computing can support a virtually unlimited set of applications,
this analysis is focused on a subset: the basic office productivity tasks that
Google Enterprise supports from the cloud. Google Enterprise3 is a suite of
services that includes:
• Google Apps – Familiar web-based applications including email, calendaring,
spreadsheets, documents, and other applications
• Vault services – Solutions for mail security, archiving, and encryption
• Enterprise Search – Google-powered search for your website or intranet
• Earth and Maps – Tools to visually display your organization’s information
• Chromebooks – Fast, lightweight computers designed to deliver the
power of the web
For concreteness, in this paper we focus on the benefits of Google Apps.
Most companies host these applications locally today. Dedicated servers on
a company’s internal network handle email, for example, while employees’
computers run standard office applications like word processing. When
a company migrates to Google Apps, shared resources in Google’s data
centers do the “number crunching” associated with these basic tasks. This lets
companies run fewer servers, which can lead to potentially large energy and
pollution savings.
Why is the cloud more energy efficient?
Energy impact
Option 1: traditional solution Option 2: cloud-based solution of the cloud

Servers

Large number of servers with low Smaller number of cloud servers

utilization and (often) sub-optimal with high utilization and efficiency
efficiency

Cooling

Inefficient, small scale cooling Advanced, continuously

without in-house professionals optimized, and highly efficient
cooling systems

Network

Significant network traffic Small increase in network traffic

The chart above compares the energy needed to support two models of office computing: the
standard model and the cloud-based model enabled by Google Apps. Migration to Google Apps
affects energy consumption in three ways:
1. Reduces direct energy for servers by 70–90%.4 Operations now require far fewer servers—and
Google’s servers are more fully loaded and significantly more efficient. A company that hosts
its own IT services must have redundancy to guard against failure and excess server capacity to
handle spiky demand. This can result in many times more servers than needed and it’s common
for companies to have server utilizations of 10% or less.5 Because servers draw nearly the same
amount of energy regardless of how busy they are, this results in a large waste of energy. Google
is able to substantially increase utilization by aggregating demand across thousands of customers.
In addition, Google data centers use customized, high-efficiency servers, power supplies, and
software that need significantly less energy per unit of output than the servers deployed at most
customer locations.6
2. Reduces energy for server cooling by 70–90%. The direct energy consumption from servers is
only part of the story. Almost every unit of energy that a server consumes ends up as heat in the
building that contains it—causing air-conditioning systems to work harder than they otherwise
would. In many situations (e.g., servers in closets or small rooms), each watt of direct power
consumption necessitates 1.5 additional watts of cooling. Even the top tier of large corporate
data centers consume ~0.5 watts of indirect power for each watt of direct power consumption.
Through significant R&D investments and innovative technologies, Google’s data centers need just
0.13 watts of cooling for each watt of direct power7—a 75–90% improvement.8 Our calculations,
which we describe in the appendix, suggest that a typical customer will see energy costs for server
cooling decrease by 70–90%.
3. Increases energy 2–3% from use of Google servers and more network traffic. While the
savings from items 1–2 are impressive, they don’t (quite) come for free. Using cloud-based
services results in some additional energy consumption from the use of Google servers and
an increase in traffic on the Internet.9 Our analysis suggests a net increase of 2–3% in energy
consumption for office computing migrating to the cloud.

2
How much can you reduce energy use and pollution by migrating
to the cloud?
The table below shows the net energy savings resulting from factors 1–3 outlined in the previous
section. Overall, savings range from 68-87%. Please refer to the appendix for details of how these
numbers were calculated.

Energy consumption Energy consumption after

Source of energy today (kWh per employee Google Apps (kWh per Savings from switching
consumption per year) employee per year) to Google Apps
Server direct energy 18–175 2–53 70–90%
Server cooling 18–88 2–26 70–90%
Total energy consumed 36–263 4–79 70–90%
per user
Additional cloud-based -- 1–5 2–3% increase
energy consumption
(Google + networking)10
Total required energy 36–263 5–84 68–87%

Savings will tend to be higher for the following segments:

• Smaller companies, which typically have very low server utilization rates and high PUEs11
• Companies whose digital workloads are heavily weighted towards the applications that
Google Apps provides (e.g., email, calendaring, spreadsheets, documents, etc.)
• Companies that make a concerted push not just to implement Google Apps in software, but also
to migrate as much as possible towards more power efficient computers (e.g., Chromebooks)
Note that we do not consider savings from switching to cloud-based client devices such as
Chromebooks. With the growth of cloud based applications, these faster, lighter and less
power-hungry devices could replace laptops for many users—leading to additional energy savings
of 10–45% for employee computers.

A real-world case study: GSA

The U.S. General Services Administration (GSA) migrated to Google Apps in 2012, and shared detailed
data about their IT equipment before and after the migration. GSA is an independent agency of the
federal government that helps manage and support federal agencies. This includes managing over
8,300 owned and leased buildings and a 210,000 vehicle motor pool. The following summary tallies
up every GSA server dedicated to email and collaboration across 14 locations in the continental
U.S. and applies the appropriate PUEs, electricity prices, and carbon intensities for each location:

Before Google Apps After Google Apps Savings

(16,742 users) (17,671 users) per user
Total number of servers operated by 324 61 82%
GSA for email and collaboration
Total direct power of GSA servers (kW) 163 22 87%
Annual GSA server energy consumption 175 20 89%
per user, direct and indirect (kWh/user)12
Additional cloud-based energy consumption -- 1-5 2–3% increase
(Google + network) (kWh/user)13
Annual GSA server energy costs ($) $307,400 $22,400 93%
Annual carbon emissions from server 1860 290 85%
energy (metric tonnes of CO2)

3
Three points stand out from these results.
1. As anticipated, by migrating to the cloud, GSA achieved significant energy savings (i.e., 89% fewer
kWh and 93% reduction in costs) and reduced pollution (i.e., 85% fewer metric tonnes of CO2), in
addition to savings from reduced server purchases and maintenance. These savings are slightly
higher than our estimates above for a typical organization, reflecting the fact that GSA’s mix of
work is well suited for a migration to Google Apps.
2. The reduction in energy consumption was larger, in percentage terms, than the reduction in
number of servers. This was due to a combination of two factors: lower requirements for average
server power under Google Apps, and consolidating GSA servers into a smaller subset of
specialized locations that had the best PUEs.
3. These results only covered energy consumption for servers and the associated cooling—not
energy consumption for client devices (e.g., laptops). The estimated direct + indirect energy
consumption of GSA’s 18,300 laptops (~3,600,000 kWh) did not significantly change during the
migration. Including this amount in the comparison, we would estimate a total client + server
energy savings of 39%.14

Conclusions
This whitepaper suggests that migrating to the cloud can produce estimated energy savings of
68–87% for typical companies. Replicating these savings across the entire economy would have a
substantial impact. According to a recent study by the Carbon Disclosure Project, by migrating to
cloud computing, large U.S. companies could achieve annual energy savings of $12.3 billion and
carbon reductions of 85.7 million metric tonnes by 2020—equivalent to the annual emissions of over
16.8 million passenger vehicles.15 This suggests that cloud computing can in fact be seen as part of
the solution towards a more energy efficient future.

4
Appendix: Notes on assumptions and methodology
This section provides justification for the numbers shown in the first table of the document.

Typical energy consumption today

We calculated the low and high range for estimates in this table by considering two types of
businesses. The low estimates are appropriate for a “data-light” business (e.g., largely clerical)
that has relatively low-power computers and servers. The high estimates are appropriate for a
data-intensive business (e.g., an engineering firm) with a greater number of higher powered
computers and servers.
For client devices (e.g., desktops, laptops, etc.), we assumed 1 laptop and 0.2 desktops per user in
the low estimate, with an annual direct energy consumption of 60 kWh for the laptop and 160 kWh
for the desktop. These energy consumption numbers are appropriate for a low-end computer or
mid-range Energy Star-qualified device.16 In the high estimate, we assumed 1 laptop and 0.8 desktops
per user, with an annual direct energy consumption of 160 kWh for the laptop and 400 kWh for
the desktop. These energy consumption figures are appropriate for a typical higher-end 60 watt
laptop and a 150 watt desktop that are on for 2,600 hours per year. We multiplied both high and low
estimates by a further factor of 1.5 to account for the additional air-conditioning needed to remove
the devices’ heat.
For servers and server cooling, we assumed an average server power of 200 watts, one server
per 50 employees, and a PUE of 2.0 in the low estimate, while assuming an average server power
of 500 watts, one server per 10 employees, and a PUE of 1.5 in the high estimate. We assumed
8,766 operating hours per year in both.

Size of potential customer energy savings in migrating to Google Apps

The low-high estimates in this section are based on the same data-light and data-intensive company
archetypes defined in the previous section.
For server energy consumption after implementing Google Apps, we assumed a 90% reduction in
number of servers in the low estimate. This large reduction reflects the fact that the work in the
“data-light” business is heavily weighted towards the basic office applications that Google Apps
can replace. In the high estimate, we assumed a 70% reduction in number of servers for the data-
intensive business, reflecting the fact that this workplace has many power users who rely on custom
applications that still need on-site server support after the Apps migration. As above, we assumed a
PUE of 2.0 for the data-light business and a PUE of 1.5 for the data-intensive business.
While not included in our analysis, for the estimated savings in client devices referenced on page 3,
we assumed in the low estimate that 80% of the laptops and 80% of the desktops in the data-light
business could be replaced by Chromebooks, reflecting the relatively routine use of computing
resources that make up employees’ work in this kind of company. We assumed in the high estimate
that only 25% of the data-intensive laptops and 5% of the data-intensive desktops could be replaced
by Chromebooks. This reflects an underlying assumption that proprietary, non-enterprise software
plays a large role in the work of employees at the data-intensive company. In both the low and high
cases, we calculated the energy savings from the replacement using the laptop and desktop energy
consumption levels described above, and a Chromebook annual energy consumption of 30 kWh/yr.
The Chromebook consumption is based on 11 watt consumption when awake, 0.5 watt consumption
when asleep, and 20% fewer “on hours” per year than the desktop or laptop it replaces. The very fast
Chromebook start-up and shut-down times encourage users to turn off the devices more frequently,
resulting in fewer “on hours.”

Per-user energy required for Google to operate Google Apps

We estimated the energy required to support one incremental Google Apps user for a year by
summing the energy required to support an average user in each of our three most popular
applications: Gmail, Calendar, and Docs (which includes spreadsheets, documents, presentations,
etc.) and then multiplying by a small scale-up factor to account for other applications provided by
Google Apps.

5
For each of the three applications, we estimated the energy requirement
per user by taking the total number of cores and disk space devoted to the
current users of that service (including redundancy), multiplying by the relevant
annual energy consumption (kWh per year per core or kWh per year per GB
storage), multiplying by Google’s trailing 12-month PUE (i.e., 1.13 at the time of
publication), and then dividing by the number of 7-day active users.
We then added a 10% energy premium per user to account for network energy
consumption, which is approximately the value advocated by Baliga and
collaborators17 for applications like Google Enterprise that have low screen-
refresh rates.
This approach leads to an estimate of ~2 kWh/yr to serve the average
Google Apps customer. Heavier users will require more energy to support,
and lighter users will require less. We have adopted the range 1–5 kWh/yr as
our energy estimate to accommodate the range of behaviors seen among
typical customers.

1. From Carbon Disclosure Project and Verdantix, Cloud Computing – The IT Solution for the 21st Century, 2011.
2. The white paper was released in the Official Google Blog post, Gmail: It’s cooler in the cloud, 2011.
3. More information is available on the Google Enterprise site at www.google.com/enterprise/.
4. This is the reduction in server energy use specific to email and collaboration services after migrating to Google Apps.
Customers may continue to maintain servers onsite for tasks that cannot be outsourced to Google Apps (e.g., specialized
design software), which we have not included in our analysis.
5. From Accenture and WSP, Cloud Computing and Sustainability: the Environmental Benefits of Moving to the Cloud,
2011.
6. Further discussion of sources of inefficiency in typical server deployments can be found in the white paper,
Google’s Green Computing: Efficiency at Scale, 2011.
7. Trailing 12-month average power usage effectiveness (PUE). Google’s PUE was 1.12 in the most recent quarter (Q1 2012).
See www.google.com/about/datacenters/inside/efficiency/power-usage.html for details.
8. Additional information on the performance of Google datacenters can be found at www.google.com/about/datacenters/
inside/efficiency/power-usage.html.
9. From Jayant Baliga et al, Green Cloud Computing: Balancing Energy in Processing, Storage, and Transport, 2011.
Proceedings of the IEEE, Vol 99, p. 149.
10. This includes the incremental energy consumption from Google data centers and network traffic needed to support
Google Apps.
11. Power Usage Effectiveness. A measurement of how much indirect energy (e.g., air conditioning) is needed to support a
server’s direct energy consumption. A PUE of 2.5 means that 1.5W of indirect power is needed for each 1W of server power,
while a PUE of 1.7 means that 0.7W of indirect power is needed for each 1W of server power.
12. This assumes a PUE of 1.7 for GSA’s two largest data centers and a PUE of 2.5 for smaller, less optimized locations.
These numbers are consistent with GSA’s estimates of their data centers at two locations (i.e., one large, one small) and
with IBM’s estimates of efficiencies for strategic (i.e., high-end) and basic data centers in their 2012 report, Data Center
Operational Efficiency Best Practices.
13. This includes the incremental energy consumption in Google data centers and network energy that is needed to
support Google Apps.
14. This is the weighted average of 87% savings in server energy and ~0% savings in client energy.
15. From Carbon Disclosure Project and Verdantix, Cloud Computing – The IT Solution for the 21st Century, 2011.
16. From the Energy Star program requirements for computers, Version 5.0.
17. From Jayant Baliga et al, Green Cloud Computing: Balancing Energy in Processing, Storage, and Transport, 2011.
Proceedings of the IEEE, Vol 99, p. 149.

© 2012 Google Inc. All rights reserved. Google, DoubleClick, the Google logo, and the DoubleClick logo are trademarks of Google
Inc. All other company and product names may be trademarks of the respective companies with which they are associated.
120615