Power to the People:

Comparing Power Usage for PCs and Thin Clients in an

Office Network Environment

Stephen Greenberg, Christa Anderson, Jennifer Mitchell-Jackson

Thin Client Computing, Scottsdale, AZ

August, 2001

[Link]

1
Table of Contents:

Purpose of this Study ...................................................................................................... 3

Test Environment ............................................................................................................ 4
Methodology.................................................................................................................... 4
Test Equipment ............................................................................................................... 6
Results of Data Collection ............................................................................................... 7
Summary of Cost Savings............................................................................................. 11
Conclusions and Recommendations ............................................................................. 13
References.................................................................................................................... 14
Sources of additional information .................................................................................. 14
Appendix A: Detailed Specifications of tested equipment ............................................ 15
Appendix B. Understanding Power Levels .................................................................... 16
About The Authors ........................................................................................................ 17

2
Purpose of this Study

Over the course of this last year energy costs have been rising--not just in California, but across the
United States. Some small businesses are seeing prices in excess of 20 cents per kilowatt hour (kWh),
resulting in noticeably higher energy bills. Given this situation, it’s natural to wonder whether using low-
power devices such as thin clients1 could reduce the power bills and, if so, by how much. However,
although there is no shortage of opinions about the real power-saving capabilities of thin clients versus
PCs, there is a serious lack of hard data readily available to back up these opinions because hardware
power ratings do not reflect actual power usage.

To fill this information gap, we present a unique practical study which introduces laboratory style test
equipment into a working production IT system. Using real-world measurements of PCs, thin client
devices, Pentium-based servers, and, networking equipment, we have developed hard data about the
power requirements of these commonly used devices. We summarize and present this data in a manner
which allows the reader to easily determine how much it costs to power their existing networks, and, the
cost savings that will result if they exchange PCs for thin client devices.

Since we intended this to be a practical study, we decided to use an actual operating business office to
ensure that our readings would reflect real-world usage patterns. We measured computers and other
equipment in combinations which reflect the way that these devices are normally used. Because the cost
of electricity varies from region to region in the United States, we’ve included a table of actual power
costs for sample areas to better illustrate the true cost/savings numbers for diverse regions within the
continental U.S. A readily available, low cost and easy-to-use power meter was used for this study to
provide a specific hardware recommendation for readers seeking to reproduce these tests on their own
equipment. Finally, we’ve included additional relevant information and background in sidebars, figures
and tables, and footnotes throughout the paper.

1
A thin client is a desktop computing device that has no hard drive and no application software running
on it locally. Applications are executed on servers while the thin client simply provides the screen display
and a way to operate the keyboard and mouse. In contrast to older style “dumb” terminals, thin clients
provide access to modern graphical PC based applications and to applications on other platforms such as
UNIX™ and mainframes. In addition, thin clients can be configured and managed remotely via a network
connection.
3
Test Environment

We selected a client of Thin Client Computing, Ameriwest Insurance, Scottsdale, AZ, as the ideal test
environment. Ameriwest uses a range of insurance specific applications and standard productivity
applications such as Microsoft Office and Internet Explorer. Traditional PC desktops are used alongside
NCD ThinSTAR Windows Based Terminals (thin clients) to access a standard file server and two
redundant Windows Terminal Servers. Remote offices, and users working at home, access the system
either through a dial-up modem or via direct Internet access. The existing IT infrastructure reflects a
typical real-world configuration for Thin Client/Server Based computing systems in use today.2 This
environment allowed for simultaneous testing of Thin Clients, traditional desktop PC’s and Servers
engaged in common tasks throughout the course of one full business week. The results, therefore, can be
easily extrapolated to networks of any size.

Methodology
Our primary questions were simple:

How much less power do thin clients use than the personal computers?

What are the actual cost savings gained by using thin client devices?

To properly answer these questions it is important to take into account the servers, and other standard
equipment, that are needed to support end user systems. To measure only the desktop device does not
lead to meaningful data. Thin client devices shift the application processing away from the desktop and
onto servers that otherwise would not exist on the network. The power consumption of these servers
must be taken into account to develop an accurate picture of true costs and potential savings. We chose
to attach our power meters to logical groups of equipment as follows:

Meter 1: A pair of load balanced Windows NT terminal servers

(4) modems for incoming remote Thin Client sessions
(4) modems for access to dial-out insurance ratings services

Meter 2: A Windows NT file server

(1) 24-port, 10/100 Ethernet Switch
(2) 8-port 10Base-T Ethernet hubs

Meter 3: (2) NCD ThinSTAR 200 Windows Based Terminals

Meter 4: A PC running a mix of applications both locally and server based

In addition to the week-long readings for the grouped devices, we took separate real-time power readings
for each individual piece of equipment. This involved running the metered equipment for several minutes
and averaging its data points to get a representative value for active power. As a result we have
meaningful power usage data both for individual components, and, these common groupings of computer
equipment.

2
The term “thin client/server based computing” is the full name for this technology. Since the applications
are actually running on the server and only displayed by the client, both “thin client” and “server” are key
elements of the system.
4
We measured the grouped devices at five-minute intervals for 5 days over a standard work week, during
which time the users engaged in their normal tasks. Each logged value was an average of the power
drawn over the 5-minute period. So that we could see full power usage for all device types (see Box 1),
we left all metered devices powered on for the duration of the test period. Details on the make, model and
specifications of each piece of equipment are listed in Appendix A.

BOX 1: Measuring Power Usage and Calculating Costs

It’s not always easy for organizations to understand their own power requirements. Computer equipment
is usually engraved with the device’s voltage and current, but these values tell you very little about the
actual power requirements. To get the rated power, you need to multiply voltage (V) times current (A)
and the power factor (PF) (i.e., V*A*PF), yet power factor is almost never indicated on the equipment. If
you are able to determine a rated power, this is still just the theoretical maximum amount of power that
the power supply can draw, not the actual power requirement. A 1990 study in the Annual Review of
Energy found that nameplate ratings on personal computers typically overstate actual measured power
by factors of 2-4.3 A more recent study of other computer devices such as switches, routers and hubs
found that these types of equipment usually draw about 30% of their rated power.4 Some manufacturers
do provide a heat rating (in British Thermal Units, or Btus) from which you can calculate a more accurate
value, but this rating is still higher than measured power. For the Dell PowerEdge 2400 server, for
example, the equipment was engraved with a rating of 330 watts. When we contacted Dell and
downloaded their rack advisor program, the Btu rating was approximately 285 watts. Under test
conditions, however, the servers used approximately 141 watts. Meters like Brand Electronics’ 21-1850
can enable users to measure actual power draws and estimate energy costs. In fact, Brand’s meter will
estimate the energy costs of each piece of equipment for you if you enter your local electricity price per
kWh.

You can audit the electricity used by your computer equipment each year in several ways. The easiest is
to perform the following calculation for each group of devices in your network:

n*p*h*52 = the number of kilo-watt hours (kWh) per year for this device group

where:

n is the number of devices of that equipment type

p is the total power (in kilowatts) used by each of those devices
h is the number of hours each week that the equipment is on
52 is the number of weeks in a year

Simply add the kilowatt-hours for each type of device and then multiply the total number of kilowatt-hours
times your average electricity price to estimate the cost of your computer equipment. Electricity prices are
usually available on the rate tariffs section of your local utilities website. For further information on your
electricity rates, contact your local utility directly.

3
Nordford et al, 1990.
4
Kunz and Kistler, 1997.
5
Test Equipment
To take our measurements, we used a Brand Electronics Model 21-1850 CI power meter and data
logger.5 This is a sophisticated power measurement device with a user-friendly interface that displays
power, energy, elapsed time and several other variables such as voltage, current, power factor, and,
average cost. Accuracy of the meter is +/- 1.5% of the reading. The meter is intended to operate between
20 and 1850 watts, so readings below 20 watts were less accurate. Since pre-testing indicated that a
single NCD ThinSTAR Windows Based Terminal draws less than 20 watts, two of these devices were
connected to Meter 3 to ensure accuracy. The modems and hubs were similarly grouped for their
individual readings. The model 21-1850 CI can store 8000 wattage data points in its internal memory,
and, provides a software interface to allow unlimited data logging of all available parameters on a
Windows 9X based PC. Both methods of data collection were used. While not as precise as more
expensive meters on the market, we chose this meter for its advanced features and ease of use. The
Brand Electronics 21-1850 CI meets the requirements of being easy to use, inexpensive, readily
available, and sufficiently accurate to give a good idea of annual electricity costs. We recommend this
device to anyone who wishes to apply the techniques used in this study to their own power requirements.

The NCD ThinSTAR 200

The Brand Electronics Model 21-1850 CI

5
Additional data on power meter and data logger can be found on the Brand Electronics website,
[Link]
6
Results of Data Collection

After our testing period was over, we uploaded wattage data from the power meters and ran individual
measurements of each device. Figure 1 shows measured power for each individual piece of equipment.

Figure 1. Comparison of power usage for each device

150
141

125

109

96
100

84
WATTS

75
69

50
39

25

8 10
6

Modem Hub Thin Ethernet PC 15" File 19" Terminal

Client Switch Monitor Server Monitor Server

7
 Figure 2 displays the power recordings over the 5-day period for the grouped devices. As is apparent
from the graph, for most hours of the day the meters showed no large fluctuations. For the majority of the
test period, recorded power for the devices on each meter fluctuated only 5 to 10 watts from the average.

Figure 2: Comparison of power levels used by Equipment Groups

316 watts

300

2 Terminal Servers
w/ Modems 281 watts

250

200
File Server
hubs, switch 175 watts
Watts

150
145 watts

100

PC 72 watts

63 watts
50

2 Thin Clients 20 watts

0
Day 1 Day 5

8
 The NCD Thin Star 200 terminals drew approximately 10 watts each. There was almost no variation in
recorded power over the 5-day period. The desktop PC that we measured at the Ameriwest site drew
nearly 7 times the active power of a Thin Client terminal—69 watts versus 10 watts. (See Figure 3.)

Figure 3. Power usage of Thin Client versus PC

75
69

50
Watts

25

10

0
Thin Client Terminal PC

The weekly data for the PC also displayed very little variation. The power requirements ranged from 62 to
73 watts over the course of the week, with an average power draw of 66 watts. Active power was
approximately 69 watts, and the PC idled at 63 or 64 watts.6 Since we did not ask users to keep active
logs of their daily activities, it is not clear what specific activities caused these fluctuations. However,
because PC power usage fluctuates little once it’s turned on, these devices are often considered a
constant load.7 Because a PC’s power is relatively constant, we were able to spot check additional PCs to
see how power requirements for differently configured PCs compared.8 We measured a Pentium 4/1.5
GHz with 384 MB of memory and found that the active power over a normal workday was approximately

6
For a description of various power modes see Appendix B, “Understanding Power Levels”
7
This is consistent with the findings in Nordford et. al, 1990 and Nordman, 2000. This does not include PCs with
power management. Power management modes are beyond the scope of this study. For information on how low-
power settings affect electricity use, see Nordman et al., 1997
8
Supported by Nordford et al., 1990 and Nordman, 2000.
9
85 watts. A 1 GHz PC with 256MB of RAM used 66 watts when running applications from a terminal
server, but 85 watts when running a compute-intensive application locally.

The largest fluctuations in our metered readings occurred on the servers, during weekday evenings. The
timing of these peaks suggests that the increased power usage occurred when the servers were
executing their nightly scheduled backups. The meter with the terminal servers peaked at a max power
14 watts higher than the average power over a roughly 30 minute interval beginning at midnight each
weekday night. The meter with the file server recorded peaks as high as 24 watts above the average
power. This peak period lasted for approximately 5 hours each weekday evening. This result is
understandable given that the file server contains the internal tape drive that is used for the nightly
backups, and, the complete backup process takes several hours to complete. This was a fortunate result
of doing our study using an actual operating business- we captured additional data that would not have
been included in a lab-only test environment.

The only other notable deviation occurred on Meter 1, for only one 5-minute interval during the test
period. Since each reading is the average power of the 5-minute period, it seems likely that one of the
pieces of equipment was rebooted during this time period. This too reflects typical usage patterns- it is
very common for Windows NT based servers to be rebooted on a weekly basis.

Notes on Computer Monitors and Power Usage

Because computer monitors are used by both thin clients and PCs we chose to meter them separately
and include the results in the consolidated cost/savings figures. We based our estimates of power
requirements for monitors on the average of measurements recorded over several minutes while the
computer was active. Three CRT monitors proved to have power draws ranging from 84 to 109 watts,
which corresponds with LBNL’s assumed active power use of 85 watts for an average monitor.9

Based on the spot checks that we performed,

power usage did not appear to perfectly
correspond to the size of the monitor. This may
be because the displays were configured at
different resolutions. Our findings indicate that
higher resolution settings increase power usage,
as does color. White screens are the most
power intensive, while black backgrounds draw
the least amount of power. As an additional
point of comparison we also tested a thin client
device with an integrated 15” LCD display.
See Table 1.

9
Kawamoto et al., February 2001. We used this value of 85 for the average monitor wattage in our cost
calculations
10
Summary of Cost Savings
Given the nature of the test environment, the methodology used, and, the data collected we were able to
accurately model the total power usage costs for both PC and thin client networks of various sizes from 5
to 100 users. The annual kilowatt hour usage of each device was calculated.10 Each network size was
analyzed for the correct number of PC’s or thin clients, monitors, file servers, hubs/switches, modems and
terminal servers. The cost of powering each type of network is the sum of the annual kilowatt hours for all
devices in that network, times the average cost per kilowatt hour. Since computers and related devices
generate heat from the power consumed, there is an additional factor to consider- the cost of cooling.
Although the amount of power required for cooling is site specific, it usually takes approximately 1 unit of
electricity to move 2 units of heat.11 This premium was also included in our calculations. The savings
gained by replacing PC’s with Thin Clients is found by subtracting the difference between the total
network costs respectively. The network cost curve was calculated at ~$.10 per kWh which is at the low
end of current energy rates. Savings are shown for both ~$.10 and ~$.20 per kWh to represent the fuller
range of energy rates currently in effect nationwide. See Figure 4.

Figure 4. Annual Electricity Costs and Savings for 5-100 Users

$10,000
PC Network
$9,000

$8,000

$7,000
Thin Client Network

$6,000

$5,000 $3000 - $6000

savings
$4,000
$2250 - $4500 savings
$3,000
$1700 - $3500 savings
$2,000
$1200 - $2500 savings
$1,000
$650 - $1400 savings
$0
0 20 40 60 80 100
Number of Users

To further assist the reader in establishing more specific potential costs saving based on their geographic
location and network size we have included energy rates from several major cities in the continental U.S.
and extended the calculations to include networks of 1000 and 2500 users. See Table 3.

10
Assumes that desktop devices are powered on for 80 hours per week. See Nordman et al, January 1997.
11
Takes into consideration the seasonal and geographic variations associated with heating as well as cooling. Based
on assumptions in Koomey et al., December 9, 1999.
11
 Table 3: Electricity Costs and Savings for various U.S. Cities

# months Annual Annual Annual Annual Annual

city provider season $/kWh seasonal Average Savings for Savings for Savings for Savings for
rate $/kWh 30 Users 100 Users 1000 Users 2500 Users

Phoenix/ Arizona Public summer $0.11 5 $0.10 $683 $3,145 $31,450 $78,627
Scottsdale Service winter $0.10 7

Commonwealth summer $0.11 3.5

Chicago $0.10 $683 $3,145 $31,450 $78,627
Edison Co. winter $0.09 8.5

summer $0.21 8
Boston NStar $0.19 $1,298 $5,976 $59,760 $149,391
winter $0.14 4

San summer $0.21 6

PGE $0.17 $1,161 $5,347 $53,470 $133,665
Francisco winter $0.13 6

summer $0.21 5
Pasadena SCE $0.17 $1,161 $5,347 $53,470 $133,665
winter $0.15 7

summer $0.14 5
San Diego SDG&E $0.14 $956 $4,403 $44,030 $110,077
winter $0.13 6

summer $0.09 8
Charlotte Duke $0.09 $615 $2,831 $28,310 $70,764
winter $0.09 4

General service for small commercial business rates as of July 27, 2001.
Does not include customer or demand charges, only energy charges. Total energy
charges includes transmission, distribution, transition, decommissioning and public
purpose funds when they vary with total kWh consumed.
Arizona: rates based on electricity consumption less than 2,500 kWh per month. If
electricity consumption increases but demand does not, then rates would fall to
approximately 7.3 cents/kWh. From
[Link]
Chicago: this includes "in lieu of demand" charges for customers < 10 kW, energy
costs would be $0.04247 if > 10kW. Does not include decommissioning
adjustment, franchise costs or state taxes as these numbers were not available
without an address. These are bundled rates if you purchase from Commonwealth
Edison, from [Link]

Boston: G-1, small commercial without a demand meter, 6 month fixed rate from
supplier, from [Link]
Pasedena: from personal communication with SCE's customer service for small
businesses, GS-1 rates, July 27, 2001.
San Francisco: includes A-1 rates plus the recent energy surcharge. From
[Link].
San Diego: no demand charge, from
[Link]
Charlotte: from personal communication with Duke Energy's customer service for
small businesses, July 27, 2001.
12
Conclusions and Recommendations

Thin Clients are far more power-efficient than PCs

Our results show that PC’s use nearly 7 times the power when compared to thin clients- 69 watts versus
10 watts. Thin client devices not only use less power, they also generate less heat. When you add this
additional factor of cooling the total power usage is equal to 103.5 watts for PCs versus 15 watts for thin
clients. Some people use older PC’s as a thin client display devices to extend the useful life of the
hardware. While this may defer the up front cost of newer hardware, it brings with it the disadvantage of
paying more for power to accomplish the same task. We recommend using thin client devices wherever
possible.

Thin Client networks cost less to power than PC networks

To develop an accurate picture of the true cost savings that results from using thin clients, it was
important to take into account all of the devices that comprise a network. This included the servers that
host the applications which the thin client displays, and, the devices which are common to both types of
networks. These factors were expected to diminish the overall savings resulting from the use of thin
clients. Yet, even with these factors taken into account, we found that thin client networks cost from 30-
60% less to power. For example, a small business with 100 users can save as much as $6,000 per year.

Power savings increase as the size of the network increases

The most obvious factor here is the increased number of low power devices found in larger networks. For
each PC that is replaced by a thin client there will be an additional power savings. However, the number
of terminal servers required to support thin clients does not increase in a linear pattern: twice as many
clients does not mean twice as many terminal servers. As clusters of servers are built to support thin
client sessions, greater economies of scale are quickly realized. Departmental servers, and servers in
remote offices, becoming increasingly unnecessary. There is no need for remote servers to store files,
forward email, or host databases, when all the applications and data are stored centrally. Our findings
suggests savings of $59,760 for a 1000 users and $149,391 for 2500 users. However, we suspect that
actual measurements of large scale networks could show even greater savings.

Power costs should be included as part of Total Cost of Ownership (TCO)

Historically, the reason for using thin clients to replace PCs has been the reduced Total Cost of
Ownership (TCO) gained through centralized administration and reduced desktop maintenance costs.
Additional benefits include improved security, reduced down-time, and, more productive remote access.
The results of this study suggest that power savings should be considered another measurable
component of reduced TCO, and, one more reason to choose thin clients over PCs when possible.

Organizations can become more aware of power usage

We recommend that organizations engage in projects to measure actual power usage to calculate the
cost of powering their IT infrastructures. The best way to lower these costs is to become aware of current
usage patterns. Once the precise factors behind power costs are well understand for a particular
environment, the direct impact of cost savings options such as thin clients can be easily measured.

For more detailed studies of your power costs, contact Thin Client Computing at info@[Link]

13
References

Kawamoto, K.; Koomey J.; Nordman B.; Piette M.A.; and Brown R.E. “Electricity Used by Office
Equipment and Network Equipment in the U.S.: Detailed Report and Appendices,” Lawrence Berkeley
National Laboratory, LBNL Publication 45917, February 2001. Document available at
[Link]

Koomey J.; Kawamoto K.; Nordman B.; Piette M.A. and Brown R.E., Memorandum to Skip Laitner, EPA
Office of Atmospheric Programs, Lawrence Berkeley National Laboratory, LBNL Publication 44698,
December 9, 1999. Document available at [Link]

Koomey, J.G.; Cramer M.; Piette M.A., and Eto J.H, “Efficiency Improvements in U.S. Office Equipment:
Expected Policy Impacts and Uncertainties,” Lawrence Berkeley National Laboratory, LBNL Publication
37383, December 1995.

Kunz M. and M. Kistler. Energy consumption of network components. Basler & Hofmann and Teleinform
AG, on behalf of the Federal Office of Energy, Berne, Ch. 1997.

Mitchell-Jackson, J. “Energy Needs in an Internet Economy: A Closer Look at Data Centers,” July 2001.
Document available at [Link]

Nordman, B. Ridgehaven Building Office Equipment Assessment. Lawrence Berkeley National

Laboratory. May 19, 2000.

Nordman, B; Piette M.A.; Kinney K. and Webber C. User Guide to Power Management in PCs and
Monitors. Lawrence Berkeley National Laboratory, LBNL Publication 39466. January 1997.

Norford, L and A. Hatcher, J. Harris, J. Roturier, O. Yu. Electricity Use In Information Technologies.
Annual Review of Energy. 1990. 15:423-53.

NStar’s non-residential tariffs for Boston from [Link].

PG&E non-residential tariffs from [Link].

Woolfe, K. Computer Energy Savings. The Green Team, Lawrence Berkeley National Lab, November
2000. Document found at [Link]/ehs/wastemin/green_team/[Link].

Sources of additional information

Thin Client Computing, Scottsdale, AZ (602) 432-8649 [Link]

Network Computing Devices Inc., Mountain View, CA (800) 800-9599 [Link]

Brand Electronics, Whitefield, ME (207) 549-3401 [Link].

14
Appendix A: Detailed Specifications of tested equipment

Thin Client Devices Specifications Power Usage (Watts)

100 Mhz RISC
NCD ThinSTAR 200 10
8MB RAM, 8MB Flash ROM

Intel Pentium 266

NCD ThinSTAR 450 13
32 RAM, 16 MB Flash ROM
Servers Specifications Power Usage (Watts)
Dual P3 533,512MB RAM
Dell PowerEdge 2400 PERC SCSI RAID, (2) 9GB
141
(Terminal Server) Windows NT 4.0 Terminal Server
Edition SP5

Single P3 533, 256 MB, PERC

Dell PowerEdge 2400 RAID (3) 18GB
96
(File Server) Windows NT Server 4.0 SP6
Internal Tape Backup Drive

Single P3 800, 512MB, (2) 9GB

Dell PowerEdge 1400
SCSI, Windows 2000 Server with 72
(Lab test)
Terminal Services
Various PC Configurations Specifications Power Usage (Watts)
Pentium Pro 200
Desktop PC #1
64 MB RAM 69
(AmeriWest)
2GB IDE

Celeron 300MHz
Desktop PC #2 128 MB RAM
41
(Lab test) 4 GB IDE hard drive
Wireless Network Card

Desktop PC #3 Pentium 1 GHz

83
(Lab test) 128 MB RAM

Pentium 1.5 GHz

Desktop PC #4
384 MB RAM 85
(Lab test)
30 GB IDE Drive
Other Devices Specifications Power Usage (Watts)
3Com SuperStack 24 port 10/100
39
model 3C16464A Ethernet Switch
Pure Data Hub 8 port 10Base-T Hub 8
Multi Tech Multi Modem
56k Data Fax Modem 6
Model MT5600ZDX

15
Appendix B. Understanding Power Levels

While our study focused mainly on active power, computer equipment may have several power modes,
as stated in Koomey, J.G.; Cramer M.; Piette M.A., and Eto J.H, “Efficiency Improvements in U.S. Office
Equipment: Expected Policy Impacts and Uncertainties,” Lawrence Berkeley National Laboratory, LBNL
Publication 37383, December 1995:

Active mode: This is the power of the device when in operation. For PC CPUs, active power can vary
somewhat when different peripherals are in operation. Monitor power can also vary depending on the
image being shown.

Standby mode: This mode represents an intermediate state which attempts to conserve power with
instant recovery. The system is idle. If the device has no standby mode, this power level is equivalent to
that of the active mode.

Suspend/Power Saving mode: This mode has the lowest power level (without being off) but has a
longer recovery time than for standby.

Off mode: The power that is drawn when the device is switched off.

Note that our study did not attempt to measure Power Saving or Off modes, although we intend to cover
these modes in future work. For more information on how power saving modes can affect electricity use,
see Nordman, B; Piette M.A.; Kinney K. and Webber C. User Guide to Power Management in PCs and
Monitors. Lawrence Berkeley National Laboratory, LBNL Publication 39466. January 1997.

16
About The Authors

Steve Greenberg is the founder and President of Thin Client Computing in Scottsdale, Arizona. A leading
expert in server based computing, he has designed mission critical solutions for various Fortune 500
companies. Email Steve at: steveg@[Link]

Jennifer Mitchell-Jackson is an independent energy consultant and a research affiliate of the Lawrence
Berkeley National Laboratory. She is a leading authority on power use in data centers and is the author of
Energy Needs in an Internet Economy: A Closer Look at Data Centers. Email Jennifer at:
jennifer@[Link]

Christa Anderson is a senior contributing editor to Windows 2000 Magazine and author and co-author of
several books, including the forthcoming Automated Deployments and Scripted Management and
Mastering Windows 2000 Server, both from Sybex. Email Christa at christa@[Link]

[Link]
17