Cut Power Costs with Virtualization
Larry Lamers
Member Technical Staff – Office of the CTO
2
Agenda
The Data Center Climate Crisis
Prioritizing Power & Cooling Initiatives
The Impact of Server Consolidation
The Future of Power Management
VMware Customer Examples
3
Data Centers are Becoming More Dense
Copyright © 2006 VMware, Inc. All rights reserved.
4
Copyright © 2006 VMware, Inc. All rights reserved.
Data Center Designs are Inefficient
Power consumed = heat which must be evacuated
Hot/cold aisles are often not set up properly
Airflow redundancy is needed account for hot pockets and humidity
Heat dissipation is more expensive than power
5
Power Management is Increasingly a Driver
Why Acquired VMW
N Am EMEA A /P
Server consolidation 64% 64% 53%
Lower datacenter costs 38% 38% 38%
Server refresh cycle 31% 24% 25%
New server for new project 21% 23% 26%
Disaster recovery initiative 21% 16% 16%
Software development / testing 21% 19% 20%
High availability initiative 17% 25% 19%
New software implementation 12% 13% 22%
Storage virtualization initiative 11% 6% 18%
Power mgt initiative 10% 9% 4%
Utility computing initiative 7% 9% 9%
Peer recommendation 5% 2% 5%
Regulatory compliance 3% 4% 2%
Consultant recommendation 1% 4% 3%
N 302 226 177
Source: VMware Survey, October 2007Question: Which three factors best explain why your organization made its last x86 virtualization software acquisition?
6
Customer Impact
Energy cost is increasingly consuming budget and hard to ignore
Data centers are out of power but half full
The risk of running out of power / space is real
Lack of power and space impedes adoption of new services
Heat causes servers to fail, resulting in service interruption
Corporate initiatives and regulations around energy are emerging
7
Virtualization…
Decouples software from
underlying hardware
Encapsulates Operating
Systems and applications
into “Virtual Machines”
Virtualization: Transformational Change in Computing
A Virtual Machine
8
Copyright © 2006 VMware, Inc. All rights reserved.
Consolidation Improves Hardware Utilization
Before VMware After VMware
Virtualization enables consolidation of workloads from underutilized servers onto a single server
to safely achieve higher utilization
9
As a Capitalist...
Energy costs are skyrocketing and consuming budgets
$4.5 billion in electricity costs in U.S. in 2006*
Costs to build new data centers are “material”: $1,000 per sq ft.
Carbon offsets: $300 per server
* EPA report to Congress on Data Center Energy Efficiency, July 2007
“Over the next 5 years,
most enterprise data
centers will spend as
much on energy (power
and cooling) as they do
on hardware
infrastructure.”
Source: Gartner,
February 2007
10
As an Environmentalist…
Data Centers are huge CO2 factories
1 full rack of blade servers = 20-25 kW = peak demand of 30 homes in California
U.S. data centers = 45 billion kWh, 1.5% of total consumption*
This electricity use has more than doubled since 2000
Every server removed or powered down saves ~12.5 tons of CO2 emissions
Equivalent to taking ~1.5 cars off the road (12,000 miles @ 20 mpg) or planting 55 trees a year
* EPA report to Congress on Data Center Energy Efficiency, July 2007
11
Solving the Climate Crisis Depends on Efficiency Gains
12
As a Data Center Operator
No more power available…
Want to move your data center?
Limited floor space / capacity
No new services!
Growth is constrained
Heat = server failure = service interruption
Your job?
13
14
15
Volume Servers are Driving Power Consumption
16
Volume Servers are Driving Power Consumption
All servers = 40% of total data center electricity use
Volume servers = 85% of all server electricity use
Up from 70% in 2000
Volume server consumption has grown 17% annually since 2000
17
Virtualization
Virtualization is THE Game Changer
18
BEFORE VMware AFTER VMware
1,000
Direct attach
3000 cables/ports
200 racks
400 power whips
80
Tiered SAN and NAS
300 cables/ports
10 racks
20 power whips
Servers
Storage
Network
Facilities
Server, Storage, and Network Consolidation
19
30:1Qualcomm
20:1AIG Technology
15:1Applied Innovation
10:1National Gypsum
10:1Antares IT
10:17-Eleven
8:1State of Montana
Conseco Finance 8:1
VMware Customer Server Consolidation Ratios
20
Energy Consumption for Power & Cooling
Nameplate ratings of servers before and after consolidation:
On average servers consume 50-67% of max power capacity
Idle servers consume 30-40% of max power capacity/rating
Copyright © 2006 VMware, Inc. All rights reserved.
Type Qty Power Rating
1 CPU 300 475 W
2 CPU 500 550 W
4 CPU 200 950 W
8 CPU -- 1600 W
Type Qty Power Rating
1 CPU -- 594 W
2 CPU 38 688 W
4 CPU 38 1188 W
8 CPU 4 2000 W
21
How Does Utilization Increase Power Consumption?
15% more power
CPU Utilization Increase
22Copyright © 2006 VMware, Inc. All rights reserved.
BEFORE AFTER
= Power: $356,554
= Cooling: $445,693
= Power: $46,513
= Cooling: $58,141
Rule of thumb: ~$700 and 7,000 kWh saved per year per workload virtualized
1 CPU 300 475 W
2 CPU 500 550 W
4 CPU 200 950 W
8 CPU -- 1600 W
1 CPU -- 594 W
2 CPU 38 688 W
4 CPU 38 1188 W
8 CPU 4 2000 W
x 67% x 67%
x $0.10x $0.10
Max Power Capacity Rating
% of Max
Cost / kWh
Cost / Yr
Savings / Year = Savings: $697,593 (86%)
kW / Yr 407 kW/hr x 24 x 365 53 kW/hr x 24 x 365
Power Savings at Utility Company
23
Higher Asset Utilization Defers Data Center Expansion
Application Demand
Addressable Capacity
Available Capacity
Time
Capacity
$ 100+ M Investment for Data Center Expansion
Increased Addressable Capacity / Cubic Feet
Stretch Real Estate Utilization / Delay Expansion
Customer Defers $100M investment by 3 years through increased space utilization
24
VDI Energy Savings
Consolidate 1,000 desktops to 16 2-way quad-cores, 32GB RAM
8 users per core or 64 per server
Desktops run 12 hours per day and servers run 24 hours
Reduce power consumption by 35%!
Save $22 and 184 kWh per PC per year
$27
227,813
632,813
938
750
Server
Virtual Desktop Infrastructure
$13
106,520
297,000
14
11
Thin Client
$40
334,333
929,813
Total
518,400kWh / year
1.44 MPower / Day (watts)
$62Cost / year / user @ $0.12/kWh
150Cooling Power (watts/hour)
120Operating Power (watts/hour)
PCs
25
HW HW HW HW HW
HW HW HW HW HW
HW HW
HW HW
HW
HW
HWHW
New Levels of Flexibility / Automation
ESX 2 / VC 1PHYSICAL VMWARE INFRASTRUCTURE 3
RP1
RP2
INDUSTRY FIRSTS:Logical Resource PoolingDistributed Resource Scheduler (DRS)Distributed Power Management (DPM)
26
Distributed Power Management (DPM)
Resource Pool
Business Demand
Power Off
Consolidates workloads onto fewer servers when the cluster needs fewer resources
Places unneeded servers in standby mode
Brings servers back online as workload needs increase
Minimizes power consumption while guaranteeing service levels
No disruption or downtime to virtual machines
27
DPM Setup and Options
These features are representative of feature areas under development. Feature commitments are subject to change and must not be included in contracts, purchase orders, or sales agreements of any kind. Technical feasibility and market demand will affect final delivery.
28
How DPM Works
Recommends/initiates host power off when load is low
Considers X minutes of load history (user defined)
All VM’s on selected host are migrated to other hosts
Host is powered off
Recommends/initiates host power on when load is high
Considers X minutes of load history (user defined)
Wake on LAN packet sent to the selected host, which boots up
DRS load-balancing kicks in and VMs are migrated to this host
These features are representative of feature areas under development. Feature commitments are subject to change and must not be included in contracts, purchase orders, or sales agreements of any kind. Technical feasibility and market demand will affect final delivery.
29
Automation Drives Additional Savings & Efficiency
These features are representative of feature areas under development. Feature commitments are subject to change and must not be included in contracts, purchase orders, or sales agreements of any kind. Technical feasibility and market demand will affect final delivery.
Customers get comfortable with DRS automation quickly
Turning off low-end workloads on “nights and weekends”reduces server use by 40%
Additional labor savings from automation
30
The Mechanics Bank
Virtualizing 102 of 105 servers
Main data center ESX Servers: 2 quad dual-core and 6 repurposed two-ways
Five year “hard” savings of $1.5M - $1.7M
Total investment of $300,000
Payback of < 10 months
31
Mechanics
Electric / HVAC Costs Drill Down
0
50000
100000
150000
200000
250000
300000
350000
Electricity Costs Cooling Costs New PDU
32
Energy Efficiency at Wyse
Annual cost: $8K
Annual savings: $70K
=> 20% of total site power bill <=
Annual cost: $78K
(12 cents / KwH)
Fully automated DRS with 7,000+ VMotions in last 6 months
70 VMs on 5 x 2-CPU servers (14:1)
½ rack
2 rack mount UPS
Backup AC on standby only
700W per server = 30 MwH / yr
Cooling = 30 x 1.25 = 38MwH
TOTAL = 68 MwH
Annual Savings: 585 MwH (90%)
60 x 2-CPU servers
4 racks
16 rack mount UPSs
Backup AC running in parallel
550W per server = 290 MwH / yr
Cooling = 290 x 1.25 = 363 MwH
TOTAL = 653 MwH
After VirtualizationBefore Virtualization
33
Savings Chart
0
100
200
300
400
500
600
700
Power Cost
Before VI
After VI
34
1-800 Radiator
Situation:
Rapid growth: 3 new franchises opened per week
Out of power, A/C maxed out, racks full
“Before we knew it our computer room was almost at maximum power capacity, our computer racks were full, our switches were all used, and our air conditioning was on continuously. Rather than spend thousands on new power systems, racks, and air conditioning, we chose to leverage VMware’s product line to allow our company to keep pace with our growth.”
-- Mike Carvalho, CTO
Results:
Removed 31 physical servers out of production – 40 workloads on 9 ESX hosts
$6,000 PG&E rebate check
25-percent reduction in power and cooling costs
35
Energy Efficiency Programs
Validation of server consolidation is recognized as a calculableand impactful energy efficiency measure
Incentive range from about $150 to $300 per server removed
Paid at 8-10 cents per KwH
Max of $4 million per site (PG&E)
Currently available throughout California…
PG&E progam: www.pge.com/hightech
SoCal Edison: http://www.sce.com/RebatesandSavings/LargeBusiness/SPC/
SDG&E/Sempra: http://sdge.com/business/specializedincentives.shtml
Programs are expanding quickly…ask your energy provider!
36
Founding member of a global consortium dedicated to developing and promoting energy efficiency for data centers by:
Defining meaningful, user-centric models and metrics
Developing standards, measurement methods, best practices and technologies to improve performance against the defined metrics
Promoting the adoption of energy efficient standards, processes, measurements and technologies
Bringing together industry leaders and end users from critical segments of the data center ecosystem to develop a unified voicearound data center efficiency issues
http://www.thegreengrid.org/
The Green Grid
37
37
Board of Directors and Founding Member CompaniesBoard of Directors and Founding Member Companies……
Who’s in The Green Grid?
38
Conclusions – How Customers Win
Server consolidation and increased utilization decrease power consumption: energy savings are natural by products
Virtualization has inherent benefits such as workload mobility that allow for superior power management
Virtualization lets customers reclaim expensive data center floor space and avoid costly data center expansion
Do more with less – virtualization offers improved service levels, responsiveness and availability with a smaller energy footprint
Virtualization is the best initiative to reduce energy
39
40
TCO Calculator
41
42
Start with a Virtualization Assessment
Ideal for companies that require a business justification to usevirtualization for a server consolidation project
Illustrates:• 3-year TCO / ROI• Target workloads to virtualize
Deliverables:• ROI model• Assessment report• Server performance and utilization
• Formal presentation of findings
43
Questions?
Thank you!
Contacts:
Product Marketing - [email protected]
Office of CTO – [email protected]