Post on 03-Aug-2020
© 2009 Optimal Innovations, HyPerformix & LSI
ptimal
Innovations
Amy Spellmann, Optimal Innovations
Richard Gimarc, HyPerformix
Charles Gimarc, LSI
Rocky Mountain CMG
June 02, 2009
Green Capacity Planning:Theory and Practice
2
Agenda
� Green Capacity Planning�Methodology
�Energy Footprint Projection
� Case Study�Server Upgrade Scenario
�Virtualized Scenario
�Storage Upgrade Options
�Comparison
� Conclusion
3
Green Capacity Planning:What is it?
� Extends established methods & best practices
� Introduces an energy energy footprint projectionfootprint projection
� Views energy as a new resource for capacity planners to include in their analysis
� Focuses on cost-effective technologies & practices to manage today’s challenges
4
Green Capacity PlanningCombines Tools & Best Practices
Predictive Capacity Planning
Energy FootprintProjection
Cost & Capacity Comparison
+ =
Server Upgrade: Energy Footprint per Month
0
5,000
10,000
15,000
20,000
25,000
Ap
r 2
00
8
Ma
y 2
00
8
Ju
n 2
00
8
Jul 2
00
8
Au
g 2
00
8
Se
p 2
00
8
Oct
20
08
No
v 2
00
8
De
c 2
00
8
Jan
20
09
Fe
b 2
00
9
Ma
r 2
00
9
Ap
r 2
00
9
Ma
y 2
00
9
Ju
n 2
00
9
Jul 2
00
9
Au
g 2
00
9
Se
p 2
00
9
Oct
20
09
No
v 2
00
9
De
c 2
00
9
Jan
20
10
Fe
b 2
01
0
Ma
r 2
01
0
Ap
r 2
01
0
kW
h p
er
Mo
nth
SU:EFP-Storage (kWh) SU:EFP-Web (kWh) SU:EFP-Image (kWh) SU:EFP-CacheApp (kWh)
SU:EFP-DB (kWh) SU:EFP-CacheFill (kWh) SU:EFP-Site (kWh)
Commercial Capacity Planning Tools
Professional Services
Green CapacityPlanning
Monthly Energy Footprint Comparison
[Energy-IT + Energy-Site]
0
5,000
10,000
15,000
20,000
25,000
Apr
200
8
Ma
y 20
08
Ju
n 2
00
8
Ju
l 20
08
Au
g 2
00
8
Se
p 2
00
8
Oct 2
00
8
Nov 2
00
8
De
c 2
00
8
Ja
n 2
00
9
Fe
b 2
00
9
Mar
200
9
Apr
200
9
Ma
y 20
09
Ju
n 2
00
9
Ju
l 20
09
Au
g 2
00
9
Se
p 2
00
9
Oct 2
00
9
Nov 2
00
9
De
c 2
00
9
Ja
n 2
01
0
Fe
b 2
01
0
Mar
201
0
Apr
201
0
kW
h p
er
Mo
nth
SU:Total kWh V:Total kWh
Established Methods
New Resource to Evaluate
5
Green Capacity Planning:Why do we need it?
� By 2010, IDC expects that for every dollar of new server spending, an additional $0.70 will be needed for power and cooling
� 2% of the world’s carbon dioxide emissions come from IT equipment--the same amount of pollution as the airline industry creates – Wall Street Journal
� With the advent of high density computer equipment such as blade servers, many data centers have maxed out their power and cooling capacity - Gartner
� New data center construction projects cost $250M and up
� Google spent $600M each for four new data centers in 2007
6
Green Capacity Planning:Value
� Determines the most efficient way to meet business needs while minimizing computing, power, cooling, space requirements
� Assures SLAs are met
� Response times
� Utilization
� Evaluates potential cost & energy savings of virtualization
MinimizeMinimizeCostCost
Meet Meet SLAsSLAs
Optimize Optimize HardwareHardwareUtilizationUtilization
7
Green Capacity Planning:Factors
Cost Effective SolutionCost Effective Solution
IT Energy
Site Energy
Energy Cost
Space
Performance
IT Equipment
� Must meet SLAs
� Response times
� Utilization
� Minimize power, space, cooling
� Evaluates potential cost & energy savings of virtualization
8
Methodology:Enhanced Traditional Approach
Business Requirements: Profile & SLA
DataCenterUpgrade Plan
IT Equipment Capacity & Workload Analysis
Energy Footprint Analysis
Infrastructure Capacity Analysis
Cost Calculation: IT Equipment,
Infrastructure & Energy
Infrastructure & Energy
Loop until Optimized
Traditional
Capacity Planning
EFP
9
Green Capacity Planning Metrics
� New Inputs�Energy Footprint (calculated or measured)
�PUE
�Active Idle
�Max Power
�Cost of electricity
� Comparison Metrics�Energy Footprint Projection
�Rack Space
�Cumulative Costs
�Productivity (Trans per kWh)
In addition to traditional capacity planning
metrics, e.g., Utilization,
Response Time, etc.
10
Energy Footprint Projection
� Determine energy required by each system component� Obtain active-idle, average,
maximum power from measurements or vendor specifications
� Compute energy usage over time (e.g., monthly)
� Add “site” infrastructure energy� Overhead energy for
running components in the data center
� Includes: cooling, fans, power distribution units, lighting, etc.
11
Energy Footprint Projection:Terminology
� Terminology� Power
� Spot measurement, single point in time
� Measured in watts (W)
� Energy� Power consumption over a period of time
� Measured in kilowatt hours (kWh)
� You pay for energy, not power
� Energy estimation� Component-level power requirements (W)
� Estimate consumption over a period of time (e.g., 1 day)
Power ≠ Energy
Power x Time = Energy Usage
12
Energy Footprint Projection:Data Center Energy Usage
General Flow� Energy enters in the form of electricity� IT equipment used to store &
manipulate information� IT equipment generates heat� Site infrastructure removes generated
heat
Data Center
Electricity
Site Infrastructure
IT Equipment
Questions � What if your data center’s power
input is maximized?� How can you increase your compute
power without increasing your power draw?
� Can you improve site efficiency?� Do you really need to build a new
data center?
13
Energy Footprint Projection:Components
� Two components� IT Equipment
� Site Infrastructure
� Approach� Compute IT energy
� Estimate Site energy
� Power Usage Effectiveness
� Today’s average of 2.0
� Use to estimate Site energy
� Energy = Power X Time Site - EstimateIT - Compute
EFP(Day) = IT_kWhComputed + Site_kWhEstimated
1.0PowerIT
PowerITPowerSitePUE ≥
+=
14
Energy Footprint Projection:Component Contribution
IT Equipment [EPA2007]
Site Infrastructure [ACC2008]
Fans - 16%
Transformer - 5%
Lighting - 8%
UPS - 17%
Chillers - 54%
Network - 10%
Storage - 10%
Servers - 80%
15
EFP(Day) = IT_kWhComputed + Site_kWhEstimated
Power Usage Effectiveness:
Estimate Site_kWh using assumed PUE:
Energy Footprint Projection:Estimating EFP(Day)
Given Poweri for each IT device:
EFP(Day) = PUE x IT_kWhComputed
1111PowerPowerPowerPowerITITITIT
PowerPowerPowerPowerITITITITPowerPowerPowerPowerSiteSiteSiteSitePUEPUEPUEPUE ≥+
=
Site_kWhEstimated = IT_kWhComputed x (PUE – 1)
IT_kWhComputed = (∑Poweri) x (24 Hours/Day)
16
Energy Footprint Projection: Server Power Usage
� How do you estimate the power usage of a server?
� Power specifications for a server (W):
� Power supply
� Max Power
� Active Idle
( ) IdleW100
Util%IdleWMaxW%Power@Util +
×−=
Idle
Max
0
50
100
150
200
250
300
0% 20% 40% 60% 80% 100%
UtilizationA
vera
ge P
ow
er
(W)
� Approximation
� Server power usage scales linearly with processor utilization
� Validated across various servers ± 5% [TGG2008]
17
Energy Footprint Projection:Server Power Usage - Example
( ) IdleW100
Util%IdleWMaxW%Power@Util +
×−=
221.9Power@70% =
Assume server power usage scales linearly with CPU utilization
( ) 147100
70147254Power@70% +
×−=
147
254
0
50
100
150
200
250
300
0% 20% 40% 60% 80% 100%
Actual Load
Avera
ge P
ow
er
(W)
18
Demonstrate Green Capacity Planning
� Extend data center lifetime while supporting future business growth�Two year planning based on business projections
�Limited data center power
� Develop a capacity plan for growth/change�Upgrade/add servers
�Leverage virtualization
� Align with business requirements�Predict required infrastructure to meet SLAs
�Reduce today’s energy footprint
�Select cost-effective solution
19
Case Study:eCommerce System
� Growing eCommerce system
� Business Requirements
� 5% monthly growth
� 2-year upgrade plan to support growth & meet SLAs
� Reduce IT & site energy costs
� Optimize hardware utilization
� Evaluate virtualization as an energy saving solution
Baseline System� 24 servers
� 42U Rack space
20
Case Study:Baseline System
� TPC-W Benchmark �Published TPC-W Benchmark Result
�May 2002
�Full Disclosure Report - details
�Simulates Web-based browsing & shopping
�Provides the baseline specification �Application architecture
�IT equipment
�Workload
�Utilization Benchmark Workload� 100,000 Items
� 76,000 Users
� 9,709 WIPSb@100,000
21
Case Study:Baseline Architecture & IT
22
Case Study:Baseline Database Storage
� 12 HDD per enclosure
� Ultra320 SCSI
� 5 enclosures
� 3U rack space per enclosure
� 41 total disks
8120243, 4, 5Database
393572Backup
1351 / 0101Log
Capacity(GB)
RAIDLevel
# ofDisk
EnclosurePurpose
3U
12
Log
23
Case Study:Database Technology Upgrade
� Move to current SAS technology that will allow scaling for 2 years
� Replace original Ultra 320 SCSI storage subsystem with newer SAS (Serial Attached SCSI) technology
� Reduce power by changing from 3½” to 2½” SFF disks
� Triple storage density by moving from 12 disks in a 3U space to 24 disks in a 2U space
� Achieve higher performance to meet the demands of the system after 2 years of workload growth
� Decrease power per enclosure by 50%
12 x 3U 24 x 2U
Ultra320 SCSI
3Gb/s SAS
24
Case Study:Modeling Approach
� Utilizing HyPerformix’s Capacity Manager for Capacity Analysis
� Dell’s Datacenter Capacity Planner for Power Metrics
� Integrating results for Server/Resource Capacity, Power Requirements & Business Metrics
25
Case Study:Modeling Scenario Strategy
Server Upgrade
� 5% monthly growth for 2 years
� 70% capacity threshold
� Optimize the number of upgrades
� Dell PE 2950 for the database
� Dell PE 1850 for all other servers
Virtualization
� 5% monthly growth for 2 years
� 70% capacity threshold
� Optimize the number of upgrades
� Dell PE 2950 for the database
� Virtualize all other servers on a pool of Dell PE 1950s
� Optimize use of virtual hosts
For each scenario1. Determine infrastructure required to support
workload growth (Capacity Manager)2. Estimate energy footprint3. Calculate cost of IT equipment & energy
26
Database Storage Growth
0
4
8
12
16
20
24
28
32
36
40
Apr-
08
May-0
8
Jun-0
8
Jul-08
Aug-0
8
Sep-0
8
Oct-
08
Nov-0
8
Dec-0
8
Jan-0
9
Feb-0
9
Mar-
09
Apr-
09
May-0
9
Jun-0
9
Jul-09
Aug-0
9
Sep-0
9
Oct-
09
Nov-0
9
Dec-0
9
Jan-1
0
Feb-1
0
Mar-
10
Apr-
10
Nu
mb
er
of D
isks
Log Backup Database
Scenario - Server Upgrade:Database Storage Provisioning
� Growth Assumption:� Database storage grows at
same rate as workload (5% per month)
� Baseline Assumptions:� Use same capacity disks
� 36GB for Log & Database� 73GB for Backup
� Same disk fill factor� One RAID adapter per
enclosure
� Method:� Estimate required disk space
based on growth� Compute number of disks� Compute number of
enclosures & adapters
� Results:� Start with 4 enclosures
� Add new enclosure when disk count reaches 24
27
Scenario - Server Upgrade:Timeline
• Minimize number of server upgrade points
• Servers upgraded once per 3-month period
• Retire old servers (replace with faster more efficient servers)
28
Scenario - Server Upgrade:Infrastructure - Final Plan
• Final 2-year capacity plan
• Infrastructure supports 5% growth over 24 months
• Server upgrades grouped to reduce downtime (one upgrade per quarter)
29
Scenario - Server Upgrade:Infrastructure Components
15.6 KWh
20.6 KWh
11.2 KWh
13.3 KWh
19.7 KWh
DBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgImgImgImgStoStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgImgImgImgStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBWEBWEB
CFCApCApCApImgImgImgImgImgImgImg
WEBWEBWEBWEBWEBWEB
StoStoStoStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgImgImgImgStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgImgImgImgStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgImgImgImgStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgImgImgImgStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBWEBWEB
CFCApCApCApImgImgImgImgImgImgImg
WEBWEBWEBWEBWEBWEB
StoStoStoStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBWEBWEB
CFCApCApCApImgImgImgImgImgImgImg
WEBWEBWEBWEBWEBWEB
StoStoStoStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBWEBWEB
CFCApCApCApImgImgImgImgImgImgImg
WEBWEBWEBWEBWEBWEB
StoStoStoStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBWEBWEB
CFCApCApCApImgImgImgImgImgImgImg
WEBWEBWEBWEBWEBWEB
StoStoStoStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBWEBWEB
CFCApCApCApImgImgImgImgImgImgImg
WEBWEBWEBWEBWEBWEB
StoStoStoStoStoStoSto
DB
WEBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgImgImgImgStoStoStoStoStoSto
WEBWEBWEB
DB
WEBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgImgImgImgStoStoStoStoStoSto
WEBWEBWEB
DBWEBWEBWEBWEBWEBWEBWEBWEBWEB
CFCApCApCApImgImgImgImgImgImgImg
WEBWEBWEBWEBWEBWEB
StoStoStoStoStoStoSto
DBWEBWEBWEBWEBWEBWEBWEBWEBWEB
CFCApCApCApImgImgImgImgImgImgImg
WEBWEBWEBWEBWEBWEB
StoStoStoStoStoStoSto
Apr
2008
May 2
008
Jun 2
008
Jul 2008
Aug 2
008
Sep 2
008
Oct
2008
Nov 2
008
Dec 2
008
Jan 2
009
Feb 2
009
Mar
2009
Apr
2009
May 2
009
Jun 2
009
Jul 2009
Aug 2
009
Sep 2
009
Oct
2009
Nov 2
009
Dec 2
009
Jan 2
010
Feb 2
010
Mar
2010
Apr
2010
At end of 2-years
• 27 Servers 21 kWh / mo
• $150K 42U rack
30
Scenario - Server Upgrade:Monthly Energy Footprint
Server Upgrade: Energy Footprint per Month
0
5,000
10,000
15,000
20,000
25,000
Apr
200
8
Ma
y 2
00
8
Ju
n 2
008
Jul 20
08
Aug
20
08
Sep
200
8
Oct 2
00
8
No
v 2
008
Dec 2
00
8
Jan
20
09
Feb
200
9
Mar
200
9
Apr
200
9
Ma
y 2
00
9
Ju
n 2
009
Jul 20
09
Aug
20
09
Sep
200
9
Oct 2
00
9
No
v 2
009
Dec 2
00
9
Jan
20
10
Feb
201
0
Mar
201
0
Apr
201
0
kW
h p
er
Mo
nth
SU:EFP-Storage (kWh) SU:EFP-Web (kWh) SU:EFP-Image (kWh) SU:EFP-CacheApp (kWh)
SU:EFP-DB (kWh) SU:EFP-CacheFill (kWh) SU:EFP-Site (kWh)
Not achieving goal of reducing energy footprint
31
Case StudyModeling Scenario Strategy
Server Upgrade
� 5% monthly growth for 2 years
� 70% capacity threshold
� Optimize the number of upgrades
� Dell PE 2950 for the database
� Dell PE 1850 for all other servers
Virtualization
� 5% monthly growth for 2 years
� 70% capacity threshold
� Optimize the number of upgrades
� Dell PE 2950 for the database
� Virtualize all other servers on a pool of Dell PE 1950s
� Optimize use of virtual hosts
For each scenario1. Determine infrastructure required to support
workload growth (Capacity Manager)2. Estimate energy footprint3. Calculate cost of IT equipment & energy
32
Scenario - Virtualization:Server Utilization Results
Virtualization: Modeling Results
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Ap
r 2
00
8
May 2
008
Jun
20
08
Jul 20
08
Au
g 2
008
Se
p 2
00
8
Oct 2
00
8
Nov 2
008
De
c 2
00
8
Jan
20
09
Fe
b 2
00
9
Ma
r 2
00
9
Ap
r 2
00
9
May 2
009
Jun
20
09
Jul 20
09
Au
g 2
009
Se
p 2
00
9
Oct 2
00
9
Nov 2
009
De
c 2
00
9
Jan
20
10
Fe
b 2
01
0
Ma
r 2
01
0
Ap
r 2
01
0
Ca
pa
cit
y U
tiliza
tio
n
0
1
2
3
4
5
6
7
8
9
10
Nu
mb
er o
f Virtu
al H
os
ts
Num VHost(*) CacheApp-AD CacheApp-S CacheApp-S2 CacheFill
Database Image(*) Web(*) VHost(*)
33
Scenario - Virtualization:Infrastructure Components
15.6 KWh
7.7 KWh
6.6 KWh
4.6 KWh
7.2 KWh
Apr
2008
May 2
008
Jun 2
008
Jul 2008
Aug 2
008
Sep 2
008
Oct
2008
Nov 2
008
Dec 2
008
Jan 2
009
Feb 2
009
Mar
2009
Apr
2009
May 2
009
Jun 2
009
Jul 2009
Aug 2
009
Sep 2
009
Oct
2009
Nov 2
009
Dec 2
009
Jan 2
010
Feb 2
010
Mar
2010
Apr
2010
DBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBWEBCF
CApCApCApImgImgImgImgImgImgImgStoStoStoStoSto
DBVHVHImgImgImgImgImgImgImgStoStoStoSto
DBVHVHImgImgImgImgImgImgImgStoStoStoSto
DBVHVHImgImgImgImgImgImgImgStoStoStoSto
DBVHVHImgImgImgImgImgImgImgStoStoStoSto
DBVHVHImgImgImgImgImgImgImgStoStoStoSto
DBVHVH
StoStoStoSto
VH
DBVHVH
StoStoStoSto
VH
DBVHVH
StoStoStoSto
VH
DBVHVH
StoStoStoSto
VH
DBVHVH
StoStoStoSto
VH
DBVHVH
StoStoStoSto
VH
DBVHVH
StoStoStoSto
VH
DBVHVH
StoStoStoSto
VH
DBVHVH
StoStoStoSto
VH
DBVH
VH
StoStoStoSto
VH
StoSto
VH
DBVH
VH
StoStoStoStoStoSto
VH
DBVH
VH
StoStoStoSto
VH
StoSto
VH
Sto
DBVH
VH
StoStoStoSto
VH
StoSto
VH
Sto
DBVH
VH
StoStoStoSto
VH
StoSto
VH
Sto
DBVH
VH
StoStoStoSto
VH
StoSto
VH
Sto
DBVH
VH
StoStoStoSto
VH
StoSto
VH
Sto
DB
VH
VH
StoStoStoSto
VH
StoSto
VH
Sto
VHDB
VH
VH
StoStoStoSto
VH
StoSto
VH
Sto
VHDB
VH
VH
StoStoStoSto
VH
StoSto
VH
Sto
VH
At end of 2-years
• 6 Servers
• 7.7 kWh / mo
• $120K
• 21U rack
34
Example - Storage Upgrade:Replace HDD with SSD
� Replace Database rotating disks with SSDs
�Random workload
�Other stores (Log, Backup) are mostly sequential
� Keep same data throughput and latency levels
�Maintain SLA
� Reduce
�DataCenter footprint
�Power
�Cost (CAPEX, Operating)
� Compare Storage at month 24 only
35
Example - Storage Upgrade:Replace HDD with SSD
� Replace 4 HDD with one SSD� Similar capacity, will be sufficient to maintain performance
� Reduced space, power & cost (capex)
240120Sequential RD MB/s
10K – 30K300Random RD IOPs
50K – 100K30K – 100KSequential IOPs
800300Cost, $/disk (est)
0.15.8Idle Power, W
2.87.9Active Power, W
2.5” SFF2.5” SFFSize
128 GB36 GBCapacity
SSD (3G SATA)HDD (3G SAS)Feature
36
4 x 2U
HDD Solution
Example - Storage Upgrade:Replace HDD with SSD
17 HDD Log
17 HDD Log
21 HDD Backup
20 HDD DB1
20 HDD DB2
19 HDD DB1
19 HDD DB2
17 HDD Log
17 HDD Log
21 HDD Backup11 SSD 11 SSD
DB1 DB27 x 2U
SSD Solution
133 HDD
3395 W Active
$68,040 capex
14U rack
55 HDD
22 SSD
1579 W Active
$66,961 capex
8U rack
42%
53%
2%
43%
37
Scenario Comparison:Server Upgrade vs. Virtualization
2142Rack Space (U)
$120,000$150,000Cumulative Cost
7,700 kWh21,000 kWhEnergy Footprint
627Server Count
VirtualizationServer Upgrade
38
Scenario Comparison:Infrastructure Sprawl
Virtualization
Baseline
Server Up
grade
LAN
Database
VHosts
VHosts_1
VHosts_2
VHosts_3
VHosts_4
24 svr, 42U
15.6 kWh/mo
9,700 WIPSb
27 svr, 42U
21 kWh/mo
31,300 WIPSb
6 svr, 21U
7.7 kWh/mo
31,300 WIPSb
39
Scenario - Virtualization:Monthly Energy Footprint
Virtualization: Energy Footprint per Month
0
5,000
10,000
15,000
20,000
25,000
Ap
r 2
00
8
Ma
y 2
00
8
Ju
n 2
00
8
Ju
l 2
00
8
Au
g 2
00
8
Se
p 2
00
8
Oct 2
00
8
No
v 2
00
8
De
c 2
00
8
Ja
n 2
00
9
Fe
b 2
00
9
Ma
r 2
00
9
Ap
r 2
00
9
Ma
y 2
00
9
Ju
n 2
00
9
Ju
l 2
00
9
Au
g 2
00
9
Se
p 2
00
9
Oct 2
00
9
No
v 2
00
9
De
c 2
00
9
Ja
n 2
01
0
Fe
b 2
01
0
Ma
r 2
01
0
Ap
r 2
01
0
kW
h p
er
Mo
nth
V:EFP-Storage (kWh) V:EFP-Web (kWh) V:EFP-Image (kWh) V:EFP-CacheApp (kWh)
V:EFP-DB (kWh) V:EFP-CacheFill (kWh) V:EFP-Vhost (kWh) V:EFP-Site (kWh)
Significant reduction:
Energy Footprint (50%)
Server Upgrade: Energy Footprint per Month
0
5,000
10,000
15,000
20,000
25,000
Apr
20
08
May 2
00
8
Jun
20
08
Jul 2
008
Au
g 2
00
8
Se
p 2
008
Oct 20
08
No
v 2
008
De
c 2
008
Ja
n 2
00
9
Fe
b 2
009
Mar
20
09
Apr
20
09
May 2
00
9
Jun
20
09
Jul 2
009
Au
g 2
00
9
Se
p 2
009
Oct 20
09
No
v 2
009
De
c 2
009
Ja
n 2
01
0
Fe
b 2
010
Mar
20
10
Apr
20
10
kW
h p
er
Mo
nth
SU:EFP-Storage (kWh) SU:EFP-Web (kWh) SU:EFP-Image (kWh) SU:EFP-CacheApp (kWh)
SU:EFP-DB (kWh) SU:EFP-CacheFill (kWh) SU:EFP-Site (kWh)
40
Scenario Comparison:Rack Space
Total Rack Space per Month
0
6
12
18
24
30
36
42
Ap
r 2
00
8
Ma
y 2
00
8
Ju
n 2
00
8
Ju
l 2
00
8
Au
g 2
00
8
Se
p 2
00
8
Oct 2
00
8
No
v 2
00
8
De
c 2
00
8
Ja
n 2
00
9
Fe
b 2
00
9
Ma
r 2
00
9
Ap
r 2
00
9
Ma
y 2
00
9
Ju
n 2
00
9
Ju
l 2
00
9
Au
g 2
00
9
Se
p 2
00
9
Oct 2
00
9
No
v 2
00
9
De
c 2
00
9
Ja
n 2
01
0
Fe
b 2
01
0
Ma
r 2
01
0
Ap
r 2
01
0
Ra
ck
Sp
ac
e (
U)
SU:Total Rack Space V:Total Rack Space
Virtualization rack space @ 50%
Server Upgrade requires same rack space at end of
2-year plan
41
Scenario Comparison:Detailed Cumulative Cost
Server Upgrade: Cumulative Monthly Cost
[Energy-IT + Energy-Site + HW-Servers + HW-Storage]
$0
$40,000
$80,000
$120,000
$160,000
Ap
r 2
00
8
Ma
y 2
00
8
Ju
n 2
00
8
Ju
l 2
00
8
Au
g 2
00
8
Se
p 2
00
8
Oct 2
00
8
No
v 2
00
8
De
c 2
00
8
Ja
n 2
00
9
Fe
b 2
00
9
Ma
r 2
00
9
Ap
r 2
00
9
Ma
y 2
00
9
Ju
n 2
00
9
Ju
l 2
00
9
Au
g 2
00
9
Se
p 2
00
9
Oct 2
00
9
No
v 2
00
9
De
c 2
00
9
Ja
n 2
01
0
Fe
b 2
01
0
Ma
r 2
01
0
Ap
r 2
01
0
Virtualization: Cumulative Monthly Cost
[Energy-IT + Energy-Site + HW-Servers + HW-Storage]
$0
$40,000
$80,000
$120,000
$160,000
Ap
r 2
00
8
Ma
y 2
00
8
Ju
n 2
00
8
Ju
l 2
00
8
Au
g 2
00
8
Se
p 2
00
8
Oct 2
00
8
No
v 2
00
8
De
c 2
00
8
Ja
n 2
00
9
Fe
b 2
00
9
Ma
r 2
00
9
Ap
r 2
00
9
Ma
y 2
00
9
Ju
n 2
00
9
Ju
l 2
00
9
Au
g 2
00
9
Se
p 2
00
9
Oct 2
00
9
No
v 2
00
9
De
c 2
00
9
Ja
n 2
01
0
Fe
b 2
01
0
Ma
r 2
01
0
Ap
r 2
01
0
• Identical storage costs
• Virtualization server cost 30% lower
• Energy cost 60% less
42
Looking Ahead
� Emerging technologies are already impacting Green Capacity Planning in the DataCenter� SSD
� Reduce # of disks & power
� Multicore Processors� Higher CPU capacity in same physical & power footprint
� Dynamic Power Management� Manage power of subsystems & components relative to current load
� Virtualization� Combine multiple servers into one server
� Facilities improvements� Creative Cooling
� Fluid cooling, higher temp operation
� Decreasing PUE� Driving PUE very close to 1.0 (lights, fans, chiller)
� Climate-driven locations
43
Energy & PUE Reductions
Reducing PUE doesn’t necessarily mean reduction in energy utilization
First 3 bars all have same PUE, but different monthly KWH
Applying PUE reductions reduce energy utilization
C.S. Base = Case Study Baseline system
C.S. Virt = Case Study system, virtualized, after 24 months of upgrades
Total KWH % of C.S. Base
0%
20%
40%
60%
80%
100%
120%
140%
C.S.
Base
C.S.
Non-Virt
C.S. Virt C.S.
Base
C.S.
Non-Virt
C.S. Virt C.S.
Base
C.S.
Non-Virt
C.S. Virt
Base Base Base Cut
Lights
Cut
Lights
Cut
Lights
Cut
Lights &
Cooling
Cut
Lights &
Cooling
Cut
Lights &
Cooling
% o
f C
ase S
tud
y B
ase
line
, K
WH
/ M
onth
PUE 2.0 PUE 1.9 PUE 1.6
44
Conclusion
� Green Capacity Planning� Energy is a resource
� Introduce Energy Footprint Projection
� Expands traditional capacity planning
� Case Study� Demonstrates the methodology
� Analyzes alternative infrastructures
� Reduces energy footprint with Virtualization
� Value� Actionable metrics
� Quantifies IT and Site energy usage & costs
� Supports responsible decision making
� Bridges the gap between IT & Facilities
45
Primary Reference
This presentation is based upon a 2008 study by:
Amy Spellmann, Optimal Innovations
amy@optimalinnovations.com
Richard Gimarc, Hyperformix
rgimarc@hyperformix.com
Charles Gimarc, LSI Corporation
Charles.Gimarc@lsi.com
And a paper presented at the 2008 Computer Measurement Group conference, December 2008. The paper may be downloaded from:
http://www.optimalinnovations.com/pdf/green_capacity_planning.pdf
46
Reference
� [ACC2008] “Data Center Energy Forecast Report, Final Report, July 2008
� [EPA2007] “Report to Congress on Server and Data Center Energy Efficiency”, U.S. Environmental Protection Agency, August 2008
� [KOOM2007] “Estimating Regional Power Consumption By Servers”, Jonathan G. Koomey, LBNL Tech Note
� [SPEC2008] SPECpower_ssj2008, Feb 27, 2008, Dell PowerEdge 2950 III (Intel Xeon E5440)
� [TGG2007] “The Green Grid Data Center Power Efficiency Metrics: PUE and DCiE”, 2007, The Green Grid, Page 6
� [TGG2008] "Five Ways to Reduce Data Center Server Power Consumption“, 2008, Mark Blackburn, The Green Grid, Page 5
� [GCP2008] “Green Capacity Planning: Theory and Practice”, Spellmann Gimarc & Gimarc, CMG 2008
� [GCP-b2008] “Green Capacity Planning”, 2008, C. Gimarc, http://wikibon.org/?c=wiki&m=v&title=Green_Capacity_Planning