Tco Methodology
Content
VMware Infrastructure 3 TCO Methodology
VMware Infrastructure 3 TCO Methodology
Table of Contents
1 2 3 4 Document Overview................................................................................................. 3
1.1 A Note on Calculations..................................................................................................3
Nomenclature........................................................................................................... 4 Bartok, Inc.: A Sample Customer Implementation.................................................... 6 Server Consolidation................................................................................................ 7
4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 VMware Software ..........................................................................................................7 VMware Services ..........................................................................................................8 VMware Training ...........................................................................................................9 Server Hardware ...........................................................................................................9 Power ..........................................................................................................................11 Cooling ........................................................................................................................13 SAN Environment........................................................................................................15 Network .......................................................................................................................18 Data Center Real Estate .............................................................................................21 Disaster Recovery CapEx and OpEx ..........................................................................24 Disaster Recovery Maintenance and Provisioning......................................................24 Downtime ....................................................................................................................24 Recovery from a Disaster or Massive Failure .............................................................25 Virtual Infrastructure Installations ................................................................................27 Provisioning.................................................................................................................28 Net Present Value .......................................................................................................30 Bartok’s 3 Year TCO Reduction ..................................................................................30
5
Business Continuity................................................................................................ 24
5.1 5.2 5.3 5.4
6 7
Infrastructure Provisioning ..................................................................................... 27
6.1 6.2 7.1 7.2
Summary................................................................................................................ 30
Appendix A: References................................................................................................ 32 Appendix B: Conversions .............................................................................................. 33 Appendix C: Links ......................................................................................................... 34 Appendix D: Pricing Regressions .................................................................................. 35
2
VMware Infrastructure 3 TCO Methodology
1 Document Overview
This document provides a methodology for capturing the reduction in Total Cost of Ownership (TCO) after implementing VMware Infrastructure 3.1 It does so through quantifying both cost and labor hour savings (or investments) across solution areas and cost categories. These categories are organized as follows: • Server Consolidation o VMware Software o VMware Services o VMware Training o Server Hardware o Power o Cooling o SAN Environment o Network o Data Center Real Estate • Business Continuity o Disaster Recovery CapEx and OpEx o Disaster Recovery Maintenance and Provisioning o General Downtime o Recovery from Disaster or Massive Failure • Infrastructure Provisioning o VMware Infrastructure 3 Installations o Workload Provisioning This document provides the basis for a detailed TCO calculator that accounts for each of the above categories of TCO reduction with respect to VMware Infrastructure 3.
1.1 A Note on Calculations
Within each section of this document, a “TCO Calculation” is provided. This is the canonical formula for achieving the “Savings” by using VMware. As such, the formula subtracts the “After VMware” scenario from the “Before VMware” scenario to achieve the change in TCO. Example calculations of the fictitious company Bartok, Inc. are often a slight abstraction of the canonical formula, and usually provide individual calculations for “Before” and “After” scenarios without illustrating the overall savings.
1
© VMware, Inc. All rights reserved. Protected by one or more of U.S. Patent Nos. 6,397,242, 6,496,847, 6,704,925, 6,711,672, 6,725,289, 6,735,601, 6,785,886, 6,789,156, 6,795,966, 6,880,022, 6,961,941, 6,961,806, 6,944,699, 7,069,413; 7,082,598 and 7,089,377; patents pending.
VMware, the VMware “boxes” logo and design, Virtual SMP and VMotion are registered trademarks or trademarks of VMware, Inc. in the United States and/or other jurisdictions. All other marks and names mentioned herein may be trademarks of their respective companies.
3
VMware Infrastructure 3 TCO Methodology
2 Nomenclature
To describe TCO reduction, several formulas are provided within each section of this document. The table below offers a consolidated view of the common variables used in the formulas.
Variable VI$(ENT) VI$(STD) VI$(STRT) VC$ s STOTAL S$ Description Unit price of VMware Infrastructure 3 Enterprise Unit price of VMware Infrastructure 3 Standard Unit price of VMware Infrastructure 3 Starter Unit price of VirtualCenter Management Center Subscription factor – percentage of unit price that yields annual fee Number of physical servers categorized by CPU Cost per physical server categorized by CPU Default Value $5750 (list price US$) $3750 (list price US$) $1000 (list price US$) $5000 (list price US$) 21% (Gold) 25% (Platinum) Customer Input Before/After 1 CPU: $4000 / $4000 2 CPU: $6500 / $10000 4 CPU: $14000 / $23000 8 CPU: $30000 / $45000 16 CPU: $140000 / $160000 32 CPU: $275000 / $320000 3 Years 4 Before/After 1 CPU: 475W / 550W 2 CPU: 550W / 675W 4 CPU: 950W / 1150W 8 CPU: 1600W / 1900W 16 CPU: 4400W / 5200W 32 CPU: 9200W / 11000W $.0813 (Average commercial value for United States in 2005) 0.67 0.8 25% 25% Calculated from total number of servers (assume switches are fully redundant) 2 HBA’s per server that is refreshed $1,000 $5,000 $6,000 Before: customer data After: Assume 20GB per virtual machine (2 vmdk files at 10GB per file) 25% before, 100% after (all VM’s will be on the SAN) Calculated from total number
SLIFE WRATIO STTLPWR
Useful life of server (also called the refresh rate) Average number of workloads per CPU Sum of the nameplate power ratings of all computing infrastructure in the data center in kW.
E$ λ L ρ δ NSANSW NHBA P$(HBA) P$(SWITCH) P$(STORAGE) σTOTAL
Price per hour of 1 kW of electricity. Steady-state constant Cooling Load Factor -amount of power consumed by the cooling equipment for 1W of heat dissipated Airflow Redundancy Constant – redundant airflow required to cool data center Inefficiency (Humidification) Constant – redundant airflow to account for burden of humidification Number of new SAN switches per year Number of new HBAs per year Price per HBA Price per SAN switch Price per terabyte of storage SAN Storage
n NNETSW
Percent of servers attached to SAN Number of new network switches per year
4
VMware Infrastructure 3 TCO Methodology
SNIC pn P$(NETSW) A$ RSF RSPACE RTOTAL r D$ τU kU τR kR IESX tESX WPROV tPROV kPT
Number of NIC’s per server Number of ports per NIC Price per SAN switch Cost per square foot to build data center Square feet per rack Percent of space used by racks Total number of racks Annual Percentage Rate Cost of downtime per hour Unplanned downtime per year Unplanned downtime reduction factor Time for full recovery after disaster Recovery reduction factor Number of ESX Installations Hours per ESX Installation Total workloads provisioned per year Total hours to provision a workload Provisioning time reduction factor – factor by which provisioning time is reduced
of servers Before: 2 per server After: 3 per server 1 $4,000 $600 per square foot 7 Square Feet (approx) 30% Supplied by customer or calculated. See below. 6.00% $20,000 (Placeholder for model – this is typically customer dependent) 15 hours 25% 40 hours (Placeholder for model – this is typically customer dependent) 25% Number of servers after virtualization 2 NA 20 hours (without VMware) 1 hour (with VMware) 13.48 hours
5
VMware Infrastructure 3 TCO Methodology
3 Bartok, Inc.: A Sample Customer Implementation
The document references a fictitious customer, Bartok, Inc., that is used to illustrate the details of each TCO category. Identical inputs are used in different examples to promote continuity of the example and the document. Bartok has environment with the following server composition: # of 1 CPU Servers # of 2 CPU Servers # of 4 CPU Servers 300 500 200
The remaining characteristics of Bartok’s data center and business are left for the individual examples. The final section of this document summarizes Bartok’s TCO reduction through implementing VMware Infrastructure 3.
6
VMware Infrastructure 3 TCO Methodology
4 Server Consolidation
This section offers the methodology for calculating ROI or TCO with respect to server consolidation.
4.1 VMware Software
Implementation of server consolidation requires a onetime cost of purchasing VMware software as well as the annual support and subscription costs. VMware products customers typically implement include: • VMware Infrastructure 3 Starter, Standard, or Enterprise • VirtualCenter Management Server Licenses are sold in two processor increments. The model derives the cost of software by taking the total number of processors after virtualization and dividing by two. Depending on the type of support and subscription program, the recurring fees vary. “Gold” and “Platinum” programs are 21% and 25% of the purchase price annually, respectively (although this varies slightly if a customer purchases “Starter Edition”. The model does not account for this slight variance.) Pricing details can be found at http://www.vmware.com/pdf/vi_pricing.pdf. Calculator Notes The default implementation of the TCO calculator currently assumes that all software that is purchased is VI3 Enterprise Edition. However, the calculator allows for manual input of a different software purchase. Inputs
Input VI$(ENT) VI$(STD) VI$(STRT) VC$ s Description Unit price of VMware Infrastructure 3 Enterprise Unit price of VMware Infrastructure 3 Standard Unit price of VMware Infrastructure 3 Starter Unit price of VirtualCenter Management Center Subscription factor – percentage of unit price that yields annual fee Default Value $5750 (list price US$) $3750 (list price US$) $1000 (list price US$) $5000 (list price US$) 21% (Gold) 25% (Platinum) Source VMware Commercial List Prices VMware Commercial List Prices VMware Commercial List Prices VMware Commercial List Prices VMware Commercial List Prices
TCO Calculation Initial Purchase (Year 0):
Cost Software = (VI $( ENT ) )(# VI 3 Ent Licenses ) + (VI $( STD ) )(# VI 3 Std Licenses ) + (VI $( STRT ) )(# VI 3 Strt Licenses) + (VC$ )(# VC Licenses )
Recurring Costs:
7
VMware Infrastructure 3 TCO Methodology
⎡(VI $( ENT ) )(# VI 3 Ent Licenses) + (VI $( STD ) )(# VI 3 Std Licenses ) + ⎤ CostSubscription = s ⎢ ⎥ ⎢(VI $( STRT ) )(# VI 3 Strt Licenses) + (VC$ )(# VC Licenses ) ⎥ ⎣ ⎦
Assumptions • List prices are used for all software and support and subscription services Example Bartok, Inc. currently has 1000 servers and projects that they will be consolidated to 80 physical servers and 260 CPU’s after VMware Infrastructure 3 implementation. Since pricing is determined on a 2 CPU basis, Bartok, Inc. will require 130 VMware Infrastructure 3 Enterprise licenses. The initial cost of the software is:
CostSoftware = ($5750)(130) = $747,500
Bartok, Inc. purchases the “Platinum” support and subscription program, yielding the following annual cost:
CostSoftware ( Sub ) = (25%)($747,500) = $186,875
4.2 VMware Services
VMware and partners offer several services that should be included in the TCO calculation. These services include: • • Planning and Design (Virtualization Assessment) Deployment
Travel and materials expenses associated with services should be added to the TCO also. Planning and Design The costs for planning and design vary depending on the size of the project. The table below provides the approximate cost for specific implementations as provided by Professional Services.
Number of Servers Less than 200 servers 200 500 1000 Cost $20,000 $100,000 $175,000 $250,000
To account for variability in services pricing, a logarithmic regression is performed using the above data points. From this regression, the cost of services can be determined from the number of servers being virtualized. The regression is shown in Appendix D. Deployment Similar to planning and design costs, deployment costs also vary by the size of the project. The table below provides approximate project costs.
8
VMware Infrastructure 3 TCO Methodology
Number of Servers 500 1000
Cost $175,000 $250,000
A linear regression was performed to describe the variable cost of services based on number of servers. The regression is shown in Appendix D.
Travel and Materials Travel and materials costs will also vary depending on the size of the project. At the time of this writing, an estimate for travel and materials is not available. A number of $10,000 has been inserted into the model as an estimate.
Example Bartok, Inc. is virtualizing 1000 workloads (currently deployed on 1000 servers). Using the formulas resulting from the regressions (See appendix D), the cost of services is calculated as follows:
CostServices = 92607 ln(1000) − 393627 = $246, 078 Cost Deployment = 1100(1000) + 70000 = $158, 000 CostTravelAndMats = $10, 000
4.3 VMware Training
There are two training classes available to systems administrators. The total cost of these classes is achieved by taking the cost of the class and multiplying it by the number of administrators taking the class. Currently, the number of classes required is left as an input for the customer. Future versions of the methodology should estimate the number of administrators who will take training based on the number of servers in the environment. The cost of training is provided in the following table:
Class VI3: Install and Configure VI3: Design, Scale, and Automate Cost $2,995 $3,295
For customers who are interested in achieving “VMware Certified Professional (VCP) on VMware Infrastructure 3”, $175 should be added to the second class. (Note, the Install and Configure class is typically a prerequisite to the Design, Scale, and Automate class. Therefore, the total two-class price, with VCP included, is $6465.)
4.4 Server Hardware
Server consolidation results in significant savings on hardware purchases since the yearly requirement for new hardware is reduced dramatically. Hardware savings can be determined by comparing the number of servers purchased per year with and without virtualization. The difference in the amount of dollars spent on hardware purchases represents the TCO
9
VMware Infrastructure 3 TCO Methodology
reduction by using VMware. Dividing this number by the Server Useful Life provides the annual savings. As an input the model requires the number of servers based on number of CPU’s per machine. The model assumes that 4 virtual machines can be run per processor. This number can be overridden if desired. Furthermore, the number of servers required after virtualization can be entered manually. Inputs
Input STOTAL S$ Description Number of physical servers categorized by CPU Cost per physical server categorized by CPU Default Value N/A Before/After 1 CPU: $4000 / $4000 2 CPU: $6500 / $10000 4 CPU: $14000 / $23000 8 CPU: $30000 / $45000 16 CPU: $140000 / $160000 32 CPU: $275000 / $320000 3 Years 4 Source Customer Data Data based on average retail price for Dell, HP, and IBM Servers – validated by VMware SE’s and customer Purchase Orders.
SLIFE WRATIO
Useful life of server (also called the refresh rate) Average number of workloads per CPU
Industry average and VMware estimate VMware estimate
TCO Calculation
⋅S ⎞ ⋅S ⎞ ⎛S ⎛S SavingsHardware = ⎜ TOTAL $ ⎟ − ⎜ TOTAL $ ⎟ ⎝ S LIFE ⎠ Before ⎝ S LIFE ⎠ After
Assumptions • Depreciation and amortization costs are not included in this calculation. • Hardware refreshes occur as follows: o In the absence of VMware, hardware is typically refreshed every 3 years. o Using VMware, all hardware is purchased new at the time of implementation. Therefore, no hardware refreshes occur over the 3 year period after the initial purchase. Example Bartok, Inc. currently has 1000 servers and projects the following hardware needs after VMware Infrastructure 3 implementation:
Type 1 CPU 2 CPU 4 CPU 8 CPU Before Quantity 300 500 200 0 Price $4,000 $6,500 $14,000 $30,000 After Quantity 0 38 38 4 Price $4,000 $10,000 $23,000 $45,000
The current yearly costs for hardware are therefore:
10
VMware Infrastructure 3 TCO Methodology
CostsHardware ( Before ) =
(300)($4000) + (500)($6500) + (200)($14000) = $2, 416,667 3
With VMware, assume all new hardware is purchased at the beginning of the implementation project, and therefore all hardware costs are absorbed in year 1:
CostsHardware ( After ) = (38)($10000) + (38)($30000) + (4)($45000) = $1, 434, 000
4.5 Power
Power consumption in the data center can be categorized into two main categories: • Computing Infrastructure (IT loads): Server hardware, network switches, SAN components, etc. • Network Critical Physical Infrastructure, or NCPI (non-IT loads): transformers, uninterruptible power supplies (UPS), power wiring, fans, air conditioners, pumps, humidifiers, and lighting. [4] A complete model would account for each category of power consumption provided above. For simplicity, however, this document focuses only on the power saved from reduction in server hardware. Given this frame of reference, the cost of power is calculated by estimating the difference in power consumption of server hardware before and after virtualization. The power consumed by server hardware can be calculated by adding up the power ratings of each server in the data center. Because this number represents the maximum power used, it should be de-rated to achieve steady-state power consumption. The steady-state constant was determined empirically. According to the American Power Conversion Corporation “…the nameplate rating of most IT devices is well in excess of the actual running load by a factor of at least 33%.” [4] Forrester Research, Inc. corroborates this idea, indicating that idle x86 servers consume between 30%-40% of maximum (rated) power. [10] Power consumption is often calculated based on the form factor of the server (1U, 2U, 4U, etc). The methodology assumes that CPU correlates to U’s according to the following details: • 1 CPU = 1U • 2 CPU = 2U • 4 CPU = 4U • 8 CPU = 6U • 16 CPU = 12U • 32 CPU = 24U Inputs
Input STTLPWR Description Sum of the nameplate power ratings of all computing Default Value Before/After 1 CPU: 475W / 550W 2 CPU: 550W / 675W 4 CPU: 950W / 1150W Source Available from manufacturer’s web site (server specs). Also possible to use capacity
11
VMware Infrastructure 3 TCO Methodology
infrastructure in the data center in kW. E$ λ Price per hour of 1 kW of electricity. Steady-state constant
8 CPU: 1600W / 1900W 16 CPU: 4400W / 5200W 32 CPU: 9200W / 11000W $.0813 (Average commercial value for United States in 2005) 0.67
tools such as Dell’s Data Center Capacity Planner.2 Energy Information Administration [2] American Power Conversion [4]. On average, nameplate ratings are 33% higher than steady state load
TCO Calculation
Savings Power = λE$ (S TTLPWR )Before − (S TTLPWR ) After
[
]
Assumptions • CPU utilization will increase with VMware, and therefore power consumption will also increase. This increase in power is accounted for by increasing the power consumption per machine before and after virtualization. The exact correlation between power consumption and CPU utilization is difficult to achieve, since power consumption by processor varies by model and brand. However, using the Dell online sizing tool, it is possible to estimate the change in power consumption based on load type. Dell offers the following load categories: Idle, Average Load, I/O Intensive, and Processor Intensive. A processor intensive load may consume up to 150W more than an idle load, and 80W more than an average load according to the tool. (Note: the tool does not provide actual utilization percentages for each of the categories given.)
Example Bartok, Inc. pays $0.0813 per kWh of electricity. The power ratings of Bartok, Inc. hardware are given in the following table:
Type 1 CPU 2 CPU 4 CPU 8 CPU Before Quantity 300 500 200 0 Power Rating 475W (0.475 kW) 550W (0.55 kW) 950W (0.95 kW) 1600W (1.60 kW) After Quantity 0 38 38 4 Power Rating 550W (0.550 kW) 675W (0.675 kW) 1150W (1.150 kW) 1900W (1.90kW)
The yearly server power costs before VMware are:
CostsPower ( Before ) = (0.67)($0.0813/ kWh) ×
[(300)(0.475kW ) + (500)(0.55kW ) + (200)(0.95kW )] ⎢1day ⎥ ⎢ 1year
The yearly server power costs after VMware are:
⎡ 24h ⎤ ⎡ 365day ⎤ ⎥ = $289,878 ⎣ ⎦⎣ ⎦
2
http://www.dell.com/html/us/products/rack_advisor_test/index.html 12
VMware Infrastructure 3 TCO Methodology
CostsPower ( After ) = (0.67)($0.0813 / kWh) ×
[(38)(0.675kW ) + (38)(1.15kW ) + (4)(1.90kW )] ⎢
⎡ 24h ⎤ ⎡ 365day ⎤ ⎥⎢ ⎥ = $36, 718 ⎣1day ⎦ ⎣ 1 year ⎦
4.6 Cooling
All data center electrical equipment produces heat. Heat producing items include IT equipment (servers, switches, SAN components), power delivery systems (UPS, Power Distribution), Air Conditioning Units, Lighting, and even people. [5] For simplicity, this section focuses solely on the heat produced by server hardware. Data center design plays an important role in determining the thermal efficiency and the cost of cooling. Many data centers still employ a front-to-back layout, which positions servers in the same direction. This means the heat emission from the back of one server feeds directly into the air intake of the front of another server. A better approach is the hot-aisle/cold-aisle layout (depicted below) which mitigates the unacceptable temperature gradients associated with front-to-back layouts. [10]
Even with an optimized data center layout, as much as 25% airflow redundancy is still required. [1] In the event of a CRAC (Computer Room Air Conditioning) unit failure, airflow redundancy will continue to satisfy cooling requirements. Furthermore, many data centers have hot spots, where heat density is greater than other areas. Focused redundancy satisfies localized data center cooling requirements. [11] Beyond requirements for airflow redundancy, data centers require additional airflow to account for inefficiencies related to humidification. Humidification is required to reduce the potential for damage resulting from a static discharge. [11] Most air conditioning systems, however, induce humidity loss that is caused by the air-cooling function of the air conditioning system initiating condensation of water vapor. To maintain an acceptable humidity level, supplemental humidification is required, which creates additional load on CRAC units. [5] The additional over-sizing can be as much as 30% of the standard load. [5]
13
VMware Infrastructure 3 TCO Methodology
As a first-order approximation, the AC power consumed in the data center is completely converted to heat. Therefore, the power rating in Watts of server hardware is equal to thermal output. [5] Furthermore, according to experiments completed in the HP Laboratories, 0.8W of power is consumed by the cooling equipment for every 1W of heat dissipation in the data center (designated in this document as the Load Factor, L). This figure is confirmed by Forrester Research, Inc., which estimates that 0.5W to 1.0W of power is required to dissipate 1W of heat. [10] Given the above information, the total cost of cooling can be calculated. Inputs
Input STTLPWR Description Sum of the nameplate power ratings of all computing infrastructure in the data center in kW. Price per hour of 1 kW of electricity. Steady-state constant Default Value Before/After 1 CPU: 475W / 550W 2 CPU: 550W / 675W 4 CPU: 950W / 1150W 8 CPU: 1600W / 1900W $.0813 (Average commercial value for United States in 2005) 0.67 Source Available from manufacturer’s web site (server specs). Also possible to use capacity tools such as Dell’s Data Center Capacity Planner.3 Energy Information Administration [2] American Power Conversion [4]. On average, nameplate ratings are 33% higher than steady state load Empirically determined in HP Laboratories. [3]
E$ λ
L
ρ
δ
Cooling Load Factor amount of power consumed by the cooling equipment for 1W of heat dissipated Airflow Redundancy Constant – redundant airflow required to cool data center Inefficiency (Humidification) Constant – redundant airflow to account for burden of humidification
0.8
25%
SearchDataCenter.com [1]
25%
SearchDataCenter.com [1]
TCO Calculation
Savings Cooling =
λLE $ (1 + ρ ) (S TTLPWR )Before − (S TTLPWR ) After δ
[
]
Assumptions • This calculation does not include the amortization and maintenance costs of the power delivery and cooling systems. For explicit calculation of this value, refer to “Cost Model for Planning, Development, and Operation of a Data Center.” [3] The value was excluded here to achieve simplicity in the model.
3
http://www.dell.com/html/us/products/rack_advisor_test/index.html 14
VMware Infrastructure 3 TCO Methodology
Example The power consumed by Bartok, Inc.’s servers is repeated below for reference.
Type 1 CPU 2 CPU 4 CPU 8 CPU Before Quantity 300 500 200 0 Power Rating 475W (0.475 kW) 550W (0.55 kW) 950W (0.95 kW) 1600W (1.60 kW) After Quantity 0 38 38 4 Power Rating 550W (0.550 kW) 675W (0.675 kW) 1150W (1.15 kW) 1900W (1.90kW)
Bartok, Inc. calculates the rated power of the servers by adding all of the nameplate ratings and determines that 407 kW of steady-state power is consumed before VMware and 51.6 kW are consumed after VMware is implemented. These figures also represent the amount of heat that is emitted in Bartok, Inc.’s data center. The cost to cool this heat emission in the current environment is therefore:
⎡ 24h ⎤ ⎡ 365day ⎤ CostCooling ( Before ) = [(0.8)($0.0813/ kWh)(1.25)(1.25)][ 407kW ] ⎢ ⎥⎢ ⎥ = $362,348 ⎣1day ⎦ ⎣ 1year ⎦
Using VMware, these costs are reduced to the following:
⎡ 24h ⎤ ⎡ 365day ⎤ CostCooling ( After ) = [(0.80)($0.0813/ kWh)(1.25)(1.25)][51.6kW ] ⎢ ⎥⎢ ⎥ = $45,897 ⎣1day ⎦ ⎣ 1year ⎦
4.7 SAN Environment
The SAN architecture is likely to change significantly using VMware Infrastructure 3. Virtualization is a significant driver for moving to a shared storage infrastructure, and companies implementing VMware Infrastructure 3 are likely to invest in more shared storage. However, companies often are convinced of the benefits of shared storage even before implementing virtualization, but costs associated with implementing shared storage appear prohibitive. Server consolidation with VMware Infrastructure 3 dramatically reduces these costs by the reducing the number of SAN switches and HBAs. The reduction in switches and HBAs resulting from virtualization yields the change in total cost of ownership. A diagram of a typical SAN Architecture is depicted below. In the diagram, the “SAN Fabric” houses the switches that represent a significant portion of the reduction in TCO.
15
VMware Infrastructure 3 TCO Methodology
Inputs
Input NSANSW Description Number of new SAN switches per year Number of new HBAs per year Price per HBA Price per SAN switch Price per terabyte of storage SAN Storage Default Value Calculated from total number of servers (assume switches are fully redundant) 2 HBA’s per server that is refreshed $1,000 $5,000 $6,000 Before: customer data After: Assume 20GB per virtual machine (2 vmdk files at 10GB per file) 3 Years 25% before, 100% after (all VM’s will be on the SAN) Source Customer data
NHBA P$(HBA) P$(SWITCH) P$(STORAGE) σTOTAL
Customer data Survey of several HBA’s (Fibre Channel) from CDW4 Survey of several switches from CDW Survey of storage from CDW 20GB per VM is a conservative estimate received from a customer. Some customers report 4GB per VM. Industry average. VMware estimate. Industry data – source not available
SLIFE n
Useful life of server (also called the refresh rate) Percent of servers attached to SAN
TCO Calculation
)⎤ )⎤ ⎡ ( N )( P ⎡ ( N )( P SavingsHBA = ⎢ HBA $( HBA) ⎥ − ⎢ HBA $( HBA) ⎥ S LIFE S LIFE ⎣ ⎦ Before ⎣ ⎦ After
4
http://www.cdw.com 16
VMware Infrastructure 3 TCO Methodology
)( P )⎤ )( P )⎤ ⎡ (N ⎡ (N SavingsSANSW = ⎢ SANSW $( SANSW ) ⎥ − ⎢ SANSW $( SANSW ) ⎥ S LIFE S LIFE ⎣ ⎦ Before ⎣ ⎦ After
⎡σ P ⎤ ⎡σ P ⎤ SavingsSTORAGE = ⎢ $( STORAGE ) ⎥ − ⎢ $( STORAGE ) ⎥ ⎣ S LIFE ⎦ Before ⎣ S LIFE ⎦ After
Because we assume 2 HBA’s per server, the number of HBA’s is calculated as follows:
N HBA = 2n( STOTAL )
Similarly, the number of SAN Switches is calculated as follows. The calculation is multiplied by 2 to account for redundant switches:
N SANSW = 2n( STOTAL / 24)
Note: this formula behaves as a step function, and STOTAL / 24 should always be rounded up to the nearest whole number. For example, if there are only 6 servers in the environment, STOTAL / 24 = .25 . To achieve a whole number value of switches, we round to 1. Therefore, the total number of switches required is 2*1 = 2.
Assumptions • There are other storage savings, such as reduced cabling, less power consumed, etc. However, for simplicity reasons, these items are omitted. • SAN Switches, HBA’s, and SAN storage are refreshed at the same rate as server hardware • There are 2 HBA’s per server • SAN switches are fully redundant • For modeling purposes, a factor can be included to indicate the percentage of servers attached to the SAN. The calculations here assume all servers that are virtualized are attached. Example In the current Bartok, Inc. example, Bartok is moving from 1000 to 80 physical servers. The customer negotiates the following prices for SAN components:
P$( HBA) = $1,000 P$( SANSW ) = $5,000 P$( STORAGE ) = $6,000
First, calculate the number of networking and storage units required. Since 1000 workloads will be virtualized, SAN storage will increase according to the following calculation:
17
VMware Infrastructure 3 TCO Methodology
⎛ 20GB ⎞⎛ 1TB ⎞ Storage = 5TB + (1000Workloads) ⎜ ⎟⎜ ⎟ = 25TB ⎝ Workloads ⎠⎝ 1000GB ⎠
Assuming 2 HBA’s per server, 25% of servers connected to the SAN before, and 100% of servers connected to the SAN after, the total number of HBAs required are:
N HBA( Before ) = 2(1000)(25%) = 500 N HBA( After ) = 2(80)(100%) = 160
Assuming redundant switches, the total switches required are:
N SANSW ( Before ) = 2[(1000)(25%) / 24] = (2)(10.42) = (2)(11) = 22 N SANSW ( After ) = 2[(80)(100%) / 24) = 2(3.33) = 2(4) = 8
The network and storage savings (yearly) are now calculated as follows: HBAs:
Cost HBA( Before ) = Cost HBA( After )
(500)($1000) = $166,667 3 (160)($1000) = = $53,333 3
SAN Switches:
CostSANSW ( Before ) = CostSANSW ( After )
Storage:
(22)($5000) = $36,667 3 (8)($5000) = = $13,333 3
CostStorage ( Before ) =
CostStorage ( After )
(5)($6000) = $10,000 3 (25)($6000) = = $50,000 3
4.8 Network
The number of physical network components is dramatically reduced with Virtual Infrastructure. With fewer physical servers attached to the network, fewer switches, NIC’s and cables are required to provide server connectivity. The diagrams below illustrate this idea.
18
VMware Infrastructure 3 TCO Methodology
Virtualized Environment
Typically, the number of NIC’s per server is increase in a virtualized scenario (3 per server), although the total number of NIC’s across all servers is substantially reduced. For the TCO calculation, the reduced NIC requirements are accounted for in the price of the server hardware, since they often accompany server purchase. Therefore, this section concentrates only on the reduction in network switches. The savings in cabling are small and therefore are not recorded here. Inputs
Input NNETSW Description Number of new network switches per year Default Value Calculated from total number of servers Source Customer data
19
VMware Infrastructure 3 TCO Methodology
SNIC pn P$(NETSW)
Number of NIC’s per server Number of ports per NIC Price per SAN switch
Before: 2 per server After: 3 per server 1 $4,000
Surveyed customers Customer data – highly dependent on types of NIC’s being used. Surveyed customers and various switches at CDW
TCO Calculation
)( P )⎤ )( P )⎤ ⎡ (N ⎡ (N SavingsNETSW = ⎢ NETSW $( NETSW ) ⎥ − ⎢ NETSW $( NETSW ) ⎥ S LIFE S LIFE ⎣ ⎦ Before ⎣ ⎦ After
NNETSW is calculated as follows
N NETSW = 2( S NIC )( pn )( STOTAL ) / 24
where the denominator represents the number of ports per network switch. Note: this formula behaves as a step function, and STOTAL / 24 should always be rounded up to the nearest whole number.
Assumptions • Other savings resulting from reduced cabling and less power consumption by network switches are omitted for simplicity. • The costs of NIC’s are accounted for in the price of the servers, and are therefore not shown here. • Network switches have the same refresh rate as servers (3 years)
Example Bartok, Inc’s requirement for network switches is expected to decrease significantly after moving from 1000 to 80 physical servers. In its current data center, Bartok has on average 2 NIC’s per server and each NIC has one port. In the future state, Bartok is expecting to increase to 3 NIC’s per server, with each NIC still having one port. Furthermore, each network switch has 24 ports that are used, and costs $4,000. First, calculate the number of total switches required:
N NETSW ( Before ) = (1000)(2)(1) / 24 = 83.33 = 84 N NETSW ( After ) = (80)(3)(1) / 24 = 10
The yearly cost of switches is now calculated as follows:
Cost NETSW ( Before ) =
(84)($4000) = $112,000 3
20
VMware Infrastructure 3 TCO Methodology
Cost NETSW ( After ) =
(10)($4000) = $13,333 3
4.9 Data Center Real Estate
Savings in data center real estate are achieved by examining the potential reduction in required space after VMware Infrastructure 3 is implemented. Due to the special infrastructure (cooling, power systems, etc) required in data centers, they are often significantly more expensive to build than standard commercial properties. A data center rated at 40W per square foot costs approximately $400 per square foot. [8] At the 2009 projection of 500W per square foot, the same data center would cost $5000 per square foot to build. [8] VMware can reduce a company’s data center footprint, and prevent future construction of other data centers. The TCO calculation accounts for the yearly data center carrying costs, but does not examine the prevention of construction of other data centers. Inputs
Input A$ Description Cost per square foot to build data center Default Value $600 per square foot Source Based on size of 60W per square foot (conservative) [9] and $400/square foot for 40W/square foot data center [8] (Note: in 2005, average size was 80W/square foot) Dell PowerEdge Rack 42105 (23.94” x 39.83”) Customer approximation Calculation Depends on customer. 6% used as an estimate.
RSF RSPACE RTOTAL r
Square feet per rack Percent of space used by racks Total number of racks Annual Percentage Rate
7 Square Feet (approx) 30% Supplied by customer or calculated. See below. 6.00%
TCO Calculation
SavingsREALESTATE = [ ( MonthlyCost )(12) ]Before − [ ( MonthlyCost )(12)] After
Assumptions • The additional space required for SAN’s is small and not included here. For example, an EMC CLARiiON CX500 can hold up to 4 TB in a 4U configuration. In the example here, the extra SAN space would only occupy 1 24U rack. Example Bartok, Inc. loads 24U per rack (recall that for simplicity U’s are roughly correlated to CPU’s). Each rack requires 7 square feet of floor space, and racks take up 30% of the total floor space in the data center. The cost of the data center is approximately $600 per square foot, based on the 60W per square foot of power that is required.
5
http://www.dell.com/downloads/global/products/pedge/en/rack_system.pdf 21
VMware Infrastructure 3 TCO Methodology
First, determine the number of racks that are required before and after virtualization
RTOTAL ( Before ) = RTOTAL ( After )
(300)(1) + (500)(2) + (200)(4) = 88 Racks 24 (38)(2) + (38)(4) + (4)(8) = = 10 Racks 24
The space required for the computing infrastructure is now calculated as follows:
Re quiredSpace( General ) = Re quiredSpace( Before ) = Re quiredSpace( After ) =
( RTOTAL )( RSF ) RSPACE
(88)(7) = 2,053 ft 2 0.30
(11)(7) = 257 ft 2 0.30
The total cost of the data center before and after virtualization can now be calculated:
Cost REALESTATE ( Before ) = (2,053)($600) = $1, 232,000 Cost REALESTATE ( Before ) = (257)($600) = $154,000
To determine the monthly (and yearly) carrying costs, the total cost of the data center can be amortized over the life of the data center (assume 10 years). Bartok’s annual percentage rate on the cost of the data center is r = 6.00%, making the periodic (monthly) rate 6.00/12 or .500%. The annuity factor (the present value of $1 paid for t=120 periods) is calculated as follows:
AF =
⎞ 1 ⎛ 1 = 90.07 ⎜1 − (10)(12) ⎟ .005 ⎝ (1 + .005) ⎠
Given this annuity factor, the monthly payments are:
PaymentMonthly ( Before ) = PaymentMonthly ( Before ) =
$1, 232,000 = $13,678 90.7 $154,000 = $1,710 90.7
Given these monthly payments, the yearly costs for the data center before and after virtualization are:
22
VMware Infrastructure 3 TCO Methodology
CostYearly ( Before ) = ($13,678)(12) = $164,133 CostYearly ( After ) = ($1,710)(12) = $20,517
23
VMware Infrastructure 3 TCO Methodology
5 Business Continuity
3 offers significant savings by improving Business Continuity. These savings can be categorized into several areas: • Disaster Recovery CapEx and OpEx – The savings achieved in a company’s primary data center based on server consolidation using VMware can also be achieved in the DR site. • Disaster Recovery Maintenance and Provisioning – Maintenance and provisioning that occurs in the DR site as a result of server refreshes can be made quicker and easier. • General Downtime – VMware can dramatically reduce planned downtime associated with scheduled software upgrades, application maintenance, hardware reconfiguration, etc. In addition, it reduces unplanned downtime such as unscheduled outages due to hardware failure, application (software) failure, etc. • Recovery from a massive failure: Unsuccessful or lengthy recoveries are commonplace, and VMware Infrastructure 3 can speed recovery and enable long-term survival In addition to the items listed above, further savings can be achieved through simplified disaster recovery planning and testing. These savings are mentioned here but not quantified below.
5.1 Disaster Recovery CapEx and OpEx
Implementing VMware at the Disaster Recovery site can yield similar reductions in hardware and operations expenses that are achieved in a customer’s primary data center. These reductions are calculated in the same manner as outlined in Section 4 of this document, and are therefore omitted here.
5.2 Disaster Recovery Maintenance and Provisioning
Servers in the Disaster Recovery site require maintenance as well as provisioning. The maintenance is not quantified here, but should be mentioned as a soft cost in TCO analysis. Savings resulting from workload provisioning should be calculated using the methodology given in Section 6 of this document.
5.3 Downtime
VMware Infrastructure 3 can help reduced both planned and unplanned downtime. Planned hardware maintenance can result in a temporary disruption of business, since servers have to be powered down. VMware’s VMotion technology drastically reduces this downtime, though, by allowing IT administrators to move running virtual machines between servers. The cost benefits of planned downtime reduction are not quantified here, but should be mentioned in TCO discussions as a soft cost. Unplanned downtime is also reduced using VMware Infrastructure 3. VMotion allows administrators to proactively move virtual machines away from overloaded hosts or from servers that have experienced hardware failure. This minimizes interruption to normal business activity, and can ultimately lead to preservation of revenue. Often customers understand the cost of downtime to their business, as well as their average downtime on a yearly basis. Using these figures, it is possible to estimate the savings resulting from reduced downtime after implementing VMware. Again, only unplanned downtime is quantified below.
24
VMware Infrastructure 3 TCO Methodology
Inputs
Input D$ τU kU Description Cost of downtime per hour Unplanned downtime per year Unplanned downtime reduction Default Value Customer dependent 15 hours 75% Source Customer input Customer input VMware projection (conservative)
TCO Calculation
SavingsDowntime = D$τ U (1 − kU )
Assumptions • The unplanned downtime reduction factor is currently a conservative estimate based on discussions with VMware engineers and sales people. Empirical data should be added when made available.
Example Bartok, Inc. knows its average downtime and cost of downtime to have the following characteristics:
Category Cost of downtime (D$) Unplanned downtime per year (τU) Value $20,000 per hour 15 hours
Furthermore, they project that very conservatively VMware infrastructure 3 will allow them to cut unplanned downtime by 25%. The cost of unplanned downtime is calculated as follows:*
CostsUnplannedDowntime ( Before ) = ($20,000)(15) = $300,000 CostsUnplannedDowntime ( After ) = ($20,000)(3.75) = $75,000
* These calculations are the unfactored formulas that yielded the “Savings” formula in the TCO Calculation section above. Note that either method of cost calculation arrives at the same savings:
SavingsDowntime = ($20,000)(15)(1 − 0.25) = $225,000
5.4 Recovery from a Disaster or Massive Failure
In the event of a massive failure (disaster), VMware Infrastructure 3 can dramatically reduce recovery times. Traditional disaster recovery solutions require that recovery hardware is a
25
VMware Infrastructure 3 TCO Methodology
duplicate of production hardware. Server configuration and complex multi-step processes in these bare-metal recoveries are time-consuming and difficult. Hardware independence and single-step file recovery offered by VMware Infrastructure 3 enable improved recovery times and an expeditious return to business normalcy. The actual gains achieved using VMware Infrastructure 3 depend on the current disaster recovery infrastructure. For example, VMware Infrastructure 3 provides substantial benefits for customers who use array-based replication, and even greater benefits to customers who execute backups from tape. (Although there are other disaster recovery solutions, only these two are mentioned here for demonstration purposes). Input
Input D$ τR Description Cost of downtime per hour Time for full recovery after disaster Recovery reduction factor (% of original recovery time with virtualization) Default Value $20,000 40 hours (Placeholder for model – this is typically customer dependent) 25% Source Emerson6 Customer input
kR
VMware projection
TCO Calculation
SavingsRecov ery = D$τ R (1 − k R )
Assumptions • The customer should know their average recovery time from a disaster based on disaster recovery tests. If a customer does not perform disaster recovery testing, savings can still be projected using this methodology. • The cost of downtime includes all recovery costs in the event of a disaster (forgone revenue, labor costs, etc). • The recovery reduction factor is currently a conservative estimate based on discussions with VMware engineers and sales people. Empirical data should be added when made available. Example Through performing disaster recovery tests, Bartok, Inc discovers that it will take a minimum of 40 hours to fully recover in the event of a massive failure. Furthermore, the tests have shown that they will lose $20,000/hour of downtime during a failure. The ease of workload deployment offered by VMware Infrastructure 3 is expected to conservatively reduce their recovery times to 25% of the current time.
CostRecov ery ( Before ) = ($20,000)(40) = $800,000 CostRecov ery ( After ) = ($20,000)(10) = $200,000
6
http://www.continuitycentral.com/news02575.htm 26
VMware Infrastructure 3 TCO Methodology
6 Infrastructure Provisioning
Infrastructure provisioning involves two cost categories. The first is the cost of installing and configuring VMware Infrastructure 3 by customer administrators. The second is the cost reduction resulting from decreased provisioning time once VMware Infrastructure 3 is implemented.
6.1 Virtual Infrastructure Installations
Description Installing and configuring VMware Infrastructure 3 in the data center will require customer administrator time. This is different that the investment made in VMware Services, described in the “VMware Services”, which accounts for the labor of VMware service providers. It is expected that customer personnel will be participating in VMware implementation as well. Multiplying the average installation time by the number of installations yields the increase in labor hours. Inputs
Input IESX tESX Description Number of ESX Installations Hours per ESX Installation Default Value Number of servers after virtualization 2 Source Derived from customer input of total number of servers Survey of customers
TCO Calculation
⎛I t ⎞ ⎛I t ⎞ HoursESX INSTALL = ⎜ ESX ESX ⎟ − ⎜ ESX ESX ⎟ ⎝ S LIFE ⎠ Before ⎝ S LIFE ⎠ After
Assumptions • The above calculation assumes that customers are familiar with VMware installation. • Most customers will measure the savings (or further investment) in labor as a function of hours rather than dollars. That is because it is unlikely that the administrator headcount will be augmented as a result of virtualization. Instead, they will be deployed to other activities as necessary. Example Bartok, Inc. is ready to install ESX on the 80 servers it has recently purchased as part of the virtualization project. System administrators have gone through training and are prepared to do the installations. The hours spent on installations are calculated as follows:
HoursESX INSTALL ( Before ) = (80)(2) = 160hours
Assuming Bartok, Inc. pays $60/hour for System administrators, this equates to the following investment:
CostsESX INSTALL = (160)($60) = $9,600
27
VMware Infrastructure 3 TCO Methodology
6.2 Provisioning
Infrastructure provisioning is made significantly easier through using VMware Infrastructure 3, since it allows administrators to provision workloads from their desk without having to acquire and implement new hardware. Provisioning savings are most reliably measured in hours. The many labor hours saved will result in redeployment of administrators to value added activities. However, it is possible to associate a dollar value to provisioning savings by multiplying the hours saved by an administrator hourly wage. Inputs
Input WPROV tPROV Description Total workloads provisioned per year Total hours to provision a workload Default Value NA 20 hours (without VMware) 1 hour (with VMware) Source Derived from total servers refreshed per year Typically 20 hours without VMware, and 1 hour with VMware. This figure is an estimate based on review of customer case studies and internal VMware projections. However, the provisioning time reduction factor can also be used (see next Input). Based on a sample of 21 customers [6]
kPT
Provisioning time reduction factor – factor by which provisioning time is reduced
13.48 hours
TCO Calculation Method 1
SavingsPROVISIONING = (WPROV t PROV ) Before − (WPROV t PROV ) After
Method 2
SavingsPROVISIONING = (WPROV t PROV ) Before −
(WPROV tPROV ) After
k PT
Assumptions • The customer may not actually see the bottom line altered as a result of provisioning savings. This is because resources are likely to be deployed on other tasks, such as disaster recovery, if not spending time on provisioning servers. This still represents “savings”, however, as more value-added activities are addressed as the result of virtualization. Example 1 Recall that Bartok, Inc. currently has 1000 physical servers (workloads) and refreshes hardware every three years (333 workloads per year). Bartok’s average provisioning time is
28
VMware Infrastructure 3 TCO Methodology
20 hours. After implementing VMware, Bartok expects its provisioning time to be reduced to 1 hour per workload. This information is illustrated in the following table:
WPROV tPROV
Before VMware 333 20
After VMware 333 1
The total hours to provision workloads before and after VMware are calculated as follows:
HoursPROVISIONING ( Before ) = (333)(20) = 6660 hours HoursPROVISIONING ( After ) = (333)(1) = 333 hours
Example 2 Instead of projecting a new provisioning time of 1 hour as offered in Example 1, Bartok uses the provisioning time reduction factor of 13.48. The hours before and after installation are now:
HoursPROVISIONING ( Before ) = (333)(20) = 6660 hours HoursPROVISIONING ( After ) =
(333)(20) = 494 hours 13.48
29
VMware Infrastructure 3 TCO Methodology
7 Summary
This section provides the TCO summary of Bartok’s implementation of VMware.
7.1 Net Present Value
The calculations in the above sections yield future savings on a yearly basis. To determine the final TCO reduction, all dollar amounts are converted to present values. The model and methodology assume that savings are realized after the implementation of VMware Infrastructure 3 is complete. The length of the implementation is an input in the model, and changing it alters the net present value of the savings. The default value for implementation is 6 months, which implies that the first discount period is 1.5 years, the second is 2.5, and the third is 3.5. The diagram below illustrates this concept.
t=0
t=0.5
t=1.5
t=2.5
t=3.5
Year 1 Implementation (Year 0)
Year 2
Year 3
NPV calculations are not shown here. This document assumes the reader is familiar with discounting cash flows.
7.2 Bartok’s 3 Year TCO Reduction
The calculations in the document have been aggregated here to show Bartok’s overall TCO reduction over a 3 year period. The image below was taken from the TCO Calculator that is derived from the methodology of this document.
0 VMware Investment VMware Software VMware Services VMware Training $ (747,500) $ (414,078) $ (129,300) $ (1,290,878) $ $ $ $ Yearly Savings 1 2 (186,875) (186,875) $ $ $ $ (186,875) (186,875) $ $ $ $ 3 (186,875) (186,875) $ $ $ $ Total (1,308,125) (414,078) (129,300) (1,851,503)
Monetary Savings (Cash Flows) Server Hardware Storage Network Power and Cooling Data Center Real Estate Disaster Recovery Downtime Provisioning
$ (1,434,000) $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ (1,434,000) $
2,416,667 96,667 98,667 569,611 143,616 600,000 225,000 360,726 4,510,953
$ $ $ $ $ $ $ $ $
2,416,667 96,667 98,667 569,611 143,616 225,000 370,326 3,920,553
$ $ $ $ $ $ $ $ $
2,416,667 96,667 98,667 569,611 143,616 225,000 370,326 3,920,553
$ $ $ $ $ $ $ $ $
5,816,000 290,000 296,000 1,708,833 430,848 600,000 675,000 1,101,379 10,918,060
Total Yearly Cash Flow Discount Period Net Present Value
$ (2,724,878) $ 0 $ (2,724,878) $
4,324,078 1.00 3,793,051
$ $
3,733,678 2.00 2,872,944
$ $
3,733,678 3.00 2,520,127
$ $
9,066,557 6,461,244
30
VMware Infrastructure 3 TCO Methodology
The above calculations leveraged a 14% discount rate. Using this value, the following financial metrics were achieved: Financial Performance NPV IRR ROI Payback Period $6,461,244 140% 375% 8.32 Months
The diagram below provides a graphical representation of present value TCO reduction.
Present Value TCO Comparison over 3 Year Horizon
$12,000,000 Network VMware Training $8,000,000 Costs Provisioning Downtime Disaster Recovery $6,000,000 Data Center Real Estate Power and Cooling $4,000,000 VMware Services VMware Software Storage $2,000,000 Server Hardware
$10,000,000
$PV: Without VMware PV: With VMware
31
VMware Infrastructure 3 TCO Methodology
Appendix A: References
[1] McFarlane, Robert, “Let’s Add an Air Conditioner,” SearchDataCenter news article, published November 30, 2005. http://searchdatacenter.techtarget.com/columnItem/0,294698,sid80_gci1148906,00.html [2] Energy Information Administration, “Average Retail Price of Electricity to Ultimate Customers by End-Use Sector, by State.” http://www.eia.doe.gov/cneaf/electricity/epm/table5_6_b.html [3] Patel, Chandrakant D., Shah, Amip J., “Cost Model for Planning, Development, and Operation of a Data Center,” Internet Systems and Storage Laboratory, HP Laboratories, Palo Alto, June 9, 2005 [4] Sawyer, Richard, “Calculating Total Power Requirements for Data Centers”, American Power Conversion, 2004 [5] Rasmussen, Neil, “Calculating Total Cooling Requirements for Data Centers”, American Power Conversion, 2003 [6] “CustomerCaseStudies - OpExSavings.xls” (INTERNAL USE ONLY) [7] Disaster Recovery. (2006, July 13). In Wikipedia, The Free Encyclopedia. Retrieved July 17, 2006, from http://en.wikipedia.org/wiki/Disaster_recovery [8] Anthes, Gary, “Data Centers Get a Makeover”, Computerworld news article, published November 1, 2005. http://www.computerworld.com/databasetopics/data/datacenter/story/0,10801,97021,00.html?S KC=home97021 [9] Hughes, Ron, “The data center of the future – Part 1 – Current trends”, The Data Center Journal news article, published May17, 2005. http://www.datacenterjournal.com/News/Article.asp?article_id=315 [10] Fichera, Richard, “Power And Cooling Heat Up The Data Center”, Forrester Research, Inc. March 8, 2006 [11] Dunlap, Kevin, and Rasmussen, Neil, “The Advantages of Row and Rack-Oriented Cooling Architectures for Data Centers”, American Power Conversion, TCO
32
VMware Infrastructure 3 TCO Methodology
Appendix B: Conversions
12,000 BTU/h = 1 ton (http://en.wikipedia.org/wiki/BTU) 12,000 BTU/h = 3516 watts (http://en.wikipedia.org/wiki/Ton) 1 kilowatt = 0.283 tons of AC
33
VMware Infrastructure 3 TCO Methodology
Appendix C: Links
http://searchwinsystems.techtarget.com/originalContent/0,289142,sid68_gci1046212,00.html http://news.zdnet.com/2100-9584_22-6067896.html http://searchdatacenter.techtarget.com/columnItem/0,294698,sid80_gci1148906,00.html http://expertanswercenter.techtarget.com/eac/blog/0,295203,sid63_tax302039,00.html http://sanjose.bizjournals.com/sanjose/stories/2006/02/27/story5.html http://www.computerworld.com/databasetopics/data/datacenter/story/0,10801,97021,00.html?S KC=news97021 http://www.itjungle.com/tfh/tfh020705-story03.html http://www.dntp.com/news/pdfs/Design%20Guidelines%20for%20High%20Density%20Data%2 0Centers.pdf http://www.upsite.com/TUIpages/whitepapers/tuiheat1.0.html http://www.archinetix.com/heat-density-costs/
34
VMware Infrastructure 3 TCO Methodology
Appendix D: Pricing Regressions
The cost of services varies by the number of servers being virtualized. This cost is typically estimated on a per project basis, however, some general data points have been provided by Professional Services. The data points and corresponding regressions are provided below.
VMware Services Planning and Design # of Servers 200 $ 500 $ 1000 $ Slope 92607 Cost 100,000 175,000 250,000 Intercept -393627
# of Servers Deployment 50 $ 100 $ Slope 1100
Cost 125,000 180,000 Intercept 70000
Planning and Design Cost vs. Number of Servers $300,000 $250,000 $200,000 Cost $150,000 $100,000 $50,000 $0 200 400 600 800 1000 1200 y = 1100x + 70000 R2 = 1 Deployment Log. (Planning and Design) Linear (Deployment) y = 92607Ln(x) - 393627 R 2 = 0.9936 Planning and Design
Number of Servers
35
VMware Infrastructure 3 TCO Methodology
36
VMware Infrastructure 3 TCO Methodology
Table of Contents
1 2 3 4 Document Overview................................................................................................. 3
1.1 A Note on Calculations..................................................................................................3
Nomenclature........................................................................................................... 4 Bartok, Inc.: A Sample Customer Implementation.................................................... 6 Server Consolidation................................................................................................ 7
4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 VMware Software ..........................................................................................................7 VMware Services ..........................................................................................................8 VMware Training ...........................................................................................................9 Server Hardware ...........................................................................................................9 Power ..........................................................................................................................11 Cooling ........................................................................................................................13 SAN Environment........................................................................................................15 Network .......................................................................................................................18 Data Center Real Estate .............................................................................................21 Disaster Recovery CapEx and OpEx ..........................................................................24 Disaster Recovery Maintenance and Provisioning......................................................24 Downtime ....................................................................................................................24 Recovery from a Disaster or Massive Failure .............................................................25 Virtual Infrastructure Installations ................................................................................27 Provisioning.................................................................................................................28 Net Present Value .......................................................................................................30 Bartok’s 3 Year TCO Reduction ..................................................................................30
5
Business Continuity................................................................................................ 24
5.1 5.2 5.3 5.4
6 7
Infrastructure Provisioning ..................................................................................... 27
6.1 6.2 7.1 7.2
Summary................................................................................................................ 30
Appendix A: References................................................................................................ 32 Appendix B: Conversions .............................................................................................. 33 Appendix C: Links ......................................................................................................... 34 Appendix D: Pricing Regressions .................................................................................. 35
2
VMware Infrastructure 3 TCO Methodology
1 Document Overview
This document provides a methodology for capturing the reduction in Total Cost of Ownership (TCO) after implementing VMware Infrastructure 3.1 It does so through quantifying both cost and labor hour savings (or investments) across solution areas and cost categories. These categories are organized as follows: • Server Consolidation o VMware Software o VMware Services o VMware Training o Server Hardware o Power o Cooling o SAN Environment o Network o Data Center Real Estate • Business Continuity o Disaster Recovery CapEx and OpEx o Disaster Recovery Maintenance and Provisioning o General Downtime o Recovery from Disaster or Massive Failure • Infrastructure Provisioning o VMware Infrastructure 3 Installations o Workload Provisioning This document provides the basis for a detailed TCO calculator that accounts for each of the above categories of TCO reduction with respect to VMware Infrastructure 3.
1.1 A Note on Calculations
Within each section of this document, a “TCO Calculation” is provided. This is the canonical formula for achieving the “Savings” by using VMware. As such, the formula subtracts the “After VMware” scenario from the “Before VMware” scenario to achieve the change in TCO. Example calculations of the fictitious company Bartok, Inc. are often a slight abstraction of the canonical formula, and usually provide individual calculations for “Before” and “After” scenarios without illustrating the overall savings.
1
© VMware, Inc. All rights reserved. Protected by one or more of U.S. Patent Nos. 6,397,242, 6,496,847, 6,704,925, 6,711,672, 6,725,289, 6,735,601, 6,785,886, 6,789,156, 6,795,966, 6,880,022, 6,961,941, 6,961,806, 6,944,699, 7,069,413; 7,082,598 and 7,089,377; patents pending.
VMware, the VMware “boxes” logo and design, Virtual SMP and VMotion are registered trademarks or trademarks of VMware, Inc. in the United States and/or other jurisdictions. All other marks and names mentioned herein may be trademarks of their respective companies.
3
VMware Infrastructure 3 TCO Methodology
2 Nomenclature
To describe TCO reduction, several formulas are provided within each section of this document. The table below offers a consolidated view of the common variables used in the formulas.
Variable VI$(ENT) VI$(STD) VI$(STRT) VC$ s STOTAL S$ Description Unit price of VMware Infrastructure 3 Enterprise Unit price of VMware Infrastructure 3 Standard Unit price of VMware Infrastructure 3 Starter Unit price of VirtualCenter Management Center Subscription factor – percentage of unit price that yields annual fee Number of physical servers categorized by CPU Cost per physical server categorized by CPU Default Value $5750 (list price US$) $3750 (list price US$) $1000 (list price US$) $5000 (list price US$) 21% (Gold) 25% (Platinum) Customer Input Before/After 1 CPU: $4000 / $4000 2 CPU: $6500 / $10000 4 CPU: $14000 / $23000 8 CPU: $30000 / $45000 16 CPU: $140000 / $160000 32 CPU: $275000 / $320000 3 Years 4 Before/After 1 CPU: 475W / 550W 2 CPU: 550W / 675W 4 CPU: 950W / 1150W 8 CPU: 1600W / 1900W 16 CPU: 4400W / 5200W 32 CPU: 9200W / 11000W $.0813 (Average commercial value for United States in 2005) 0.67 0.8 25% 25% Calculated from total number of servers (assume switches are fully redundant) 2 HBA’s per server that is refreshed $1,000 $5,000 $6,000 Before: customer data After: Assume 20GB per virtual machine (2 vmdk files at 10GB per file) 25% before, 100% after (all VM’s will be on the SAN) Calculated from total number
SLIFE WRATIO STTLPWR
Useful life of server (also called the refresh rate) Average number of workloads per CPU Sum of the nameplate power ratings of all computing infrastructure in the data center in kW.
E$ λ L ρ δ NSANSW NHBA P$(HBA) P$(SWITCH) P$(STORAGE) σTOTAL
Price per hour of 1 kW of electricity. Steady-state constant Cooling Load Factor -amount of power consumed by the cooling equipment for 1W of heat dissipated Airflow Redundancy Constant – redundant airflow required to cool data center Inefficiency (Humidification) Constant – redundant airflow to account for burden of humidification Number of new SAN switches per year Number of new HBAs per year Price per HBA Price per SAN switch Price per terabyte of storage SAN Storage
n NNETSW
Percent of servers attached to SAN Number of new network switches per year
4
VMware Infrastructure 3 TCO Methodology
SNIC pn P$(NETSW) A$ RSF RSPACE RTOTAL r D$ τU kU τR kR IESX tESX WPROV tPROV kPT
Number of NIC’s per server Number of ports per NIC Price per SAN switch Cost per square foot to build data center Square feet per rack Percent of space used by racks Total number of racks Annual Percentage Rate Cost of downtime per hour Unplanned downtime per year Unplanned downtime reduction factor Time for full recovery after disaster Recovery reduction factor Number of ESX Installations Hours per ESX Installation Total workloads provisioned per year Total hours to provision a workload Provisioning time reduction factor – factor by which provisioning time is reduced
of servers Before: 2 per server After: 3 per server 1 $4,000 $600 per square foot 7 Square Feet (approx) 30% Supplied by customer or calculated. See below. 6.00% $20,000 (Placeholder for model – this is typically customer dependent) 15 hours 25% 40 hours (Placeholder for model – this is typically customer dependent) 25% Number of servers after virtualization 2 NA 20 hours (without VMware) 1 hour (with VMware) 13.48 hours
5
VMware Infrastructure 3 TCO Methodology
3 Bartok, Inc.: A Sample Customer Implementation
The document references a fictitious customer, Bartok, Inc., that is used to illustrate the details of each TCO category. Identical inputs are used in different examples to promote continuity of the example and the document. Bartok has environment with the following server composition: # of 1 CPU Servers # of 2 CPU Servers # of 4 CPU Servers 300 500 200
The remaining characteristics of Bartok’s data center and business are left for the individual examples. The final section of this document summarizes Bartok’s TCO reduction through implementing VMware Infrastructure 3.
6
VMware Infrastructure 3 TCO Methodology
4 Server Consolidation
This section offers the methodology for calculating ROI or TCO with respect to server consolidation.
4.1 VMware Software
Implementation of server consolidation requires a onetime cost of purchasing VMware software as well as the annual support and subscription costs. VMware products customers typically implement include: • VMware Infrastructure 3 Starter, Standard, or Enterprise • VirtualCenter Management Server Licenses are sold in two processor increments. The model derives the cost of software by taking the total number of processors after virtualization and dividing by two. Depending on the type of support and subscription program, the recurring fees vary. “Gold” and “Platinum” programs are 21% and 25% of the purchase price annually, respectively (although this varies slightly if a customer purchases “Starter Edition”. The model does not account for this slight variance.) Pricing details can be found at http://www.vmware.com/pdf/vi_pricing.pdf. Calculator Notes The default implementation of the TCO calculator currently assumes that all software that is purchased is VI3 Enterprise Edition. However, the calculator allows for manual input of a different software purchase. Inputs
Input VI$(ENT) VI$(STD) VI$(STRT) VC$ s Description Unit price of VMware Infrastructure 3 Enterprise Unit price of VMware Infrastructure 3 Standard Unit price of VMware Infrastructure 3 Starter Unit price of VirtualCenter Management Center Subscription factor – percentage of unit price that yields annual fee Default Value $5750 (list price US$) $3750 (list price US$) $1000 (list price US$) $5000 (list price US$) 21% (Gold) 25% (Platinum) Source VMware Commercial List Prices VMware Commercial List Prices VMware Commercial List Prices VMware Commercial List Prices VMware Commercial List Prices
TCO Calculation Initial Purchase (Year 0):
Cost Software = (VI $( ENT ) )(# VI 3 Ent Licenses ) + (VI $( STD ) )(# VI 3 Std Licenses ) + (VI $( STRT ) )(# VI 3 Strt Licenses) + (VC$ )(# VC Licenses )
Recurring Costs:
7
VMware Infrastructure 3 TCO Methodology
⎡(VI $( ENT ) )(# VI 3 Ent Licenses) + (VI $( STD ) )(# VI 3 Std Licenses ) + ⎤ CostSubscription = s ⎢ ⎥ ⎢(VI $( STRT ) )(# VI 3 Strt Licenses) + (VC$ )(# VC Licenses ) ⎥ ⎣ ⎦
Assumptions • List prices are used for all software and support and subscription services Example Bartok, Inc. currently has 1000 servers and projects that they will be consolidated to 80 physical servers and 260 CPU’s after VMware Infrastructure 3 implementation. Since pricing is determined on a 2 CPU basis, Bartok, Inc. will require 130 VMware Infrastructure 3 Enterprise licenses. The initial cost of the software is:
CostSoftware = ($5750)(130) = $747,500
Bartok, Inc. purchases the “Platinum” support and subscription program, yielding the following annual cost:
CostSoftware ( Sub ) = (25%)($747,500) = $186,875
4.2 VMware Services
VMware and partners offer several services that should be included in the TCO calculation. These services include: • • Planning and Design (Virtualization Assessment) Deployment
Travel and materials expenses associated with services should be added to the TCO also. Planning and Design The costs for planning and design vary depending on the size of the project. The table below provides the approximate cost for specific implementations as provided by Professional Services.
Number of Servers Less than 200 servers 200 500 1000 Cost $20,000 $100,000 $175,000 $250,000
To account for variability in services pricing, a logarithmic regression is performed using the above data points. From this regression, the cost of services can be determined from the number of servers being virtualized. The regression is shown in Appendix D. Deployment Similar to planning and design costs, deployment costs also vary by the size of the project. The table below provides approximate project costs.
8
VMware Infrastructure 3 TCO Methodology
Number of Servers 500 1000
Cost $175,000 $250,000
A linear regression was performed to describe the variable cost of services based on number of servers. The regression is shown in Appendix D.
Travel and Materials Travel and materials costs will also vary depending on the size of the project. At the time of this writing, an estimate for travel and materials is not available. A number of $10,000 has been inserted into the model as an estimate.
Example Bartok, Inc. is virtualizing 1000 workloads (currently deployed on 1000 servers). Using the formulas resulting from the regressions (See appendix D), the cost of services is calculated as follows:
CostServices = 92607 ln(1000) − 393627 = $246, 078 Cost Deployment = 1100(1000) + 70000 = $158, 000 CostTravelAndMats = $10, 000
4.3 VMware Training
There are two training classes available to systems administrators. The total cost of these classes is achieved by taking the cost of the class and multiplying it by the number of administrators taking the class. Currently, the number of classes required is left as an input for the customer. Future versions of the methodology should estimate the number of administrators who will take training based on the number of servers in the environment. The cost of training is provided in the following table:
Class VI3: Install and Configure VI3: Design, Scale, and Automate Cost $2,995 $3,295
For customers who are interested in achieving “VMware Certified Professional (VCP) on VMware Infrastructure 3”, $175 should be added to the second class. (Note, the Install and Configure class is typically a prerequisite to the Design, Scale, and Automate class. Therefore, the total two-class price, with VCP included, is $6465.)
4.4 Server Hardware
Server consolidation results in significant savings on hardware purchases since the yearly requirement for new hardware is reduced dramatically. Hardware savings can be determined by comparing the number of servers purchased per year with and without virtualization. The difference in the amount of dollars spent on hardware purchases represents the TCO
9
VMware Infrastructure 3 TCO Methodology
reduction by using VMware. Dividing this number by the Server Useful Life provides the annual savings. As an input the model requires the number of servers based on number of CPU’s per machine. The model assumes that 4 virtual machines can be run per processor. This number can be overridden if desired. Furthermore, the number of servers required after virtualization can be entered manually. Inputs
Input STOTAL S$ Description Number of physical servers categorized by CPU Cost per physical server categorized by CPU Default Value N/A Before/After 1 CPU: $4000 / $4000 2 CPU: $6500 / $10000 4 CPU: $14000 / $23000 8 CPU: $30000 / $45000 16 CPU: $140000 / $160000 32 CPU: $275000 / $320000 3 Years 4 Source Customer Data Data based on average retail price for Dell, HP, and IBM Servers – validated by VMware SE’s and customer Purchase Orders.
SLIFE WRATIO
Useful life of server (also called the refresh rate) Average number of workloads per CPU
Industry average and VMware estimate VMware estimate
TCO Calculation
⋅S ⎞ ⋅S ⎞ ⎛S ⎛S SavingsHardware = ⎜ TOTAL $ ⎟ − ⎜ TOTAL $ ⎟ ⎝ S LIFE ⎠ Before ⎝ S LIFE ⎠ After
Assumptions • Depreciation and amortization costs are not included in this calculation. • Hardware refreshes occur as follows: o In the absence of VMware, hardware is typically refreshed every 3 years. o Using VMware, all hardware is purchased new at the time of implementation. Therefore, no hardware refreshes occur over the 3 year period after the initial purchase. Example Bartok, Inc. currently has 1000 servers and projects the following hardware needs after VMware Infrastructure 3 implementation:
Type 1 CPU 2 CPU 4 CPU 8 CPU Before Quantity 300 500 200 0 Price $4,000 $6,500 $14,000 $30,000 After Quantity 0 38 38 4 Price $4,000 $10,000 $23,000 $45,000
The current yearly costs for hardware are therefore:
10
VMware Infrastructure 3 TCO Methodology
CostsHardware ( Before ) =
(300)($4000) + (500)($6500) + (200)($14000) = $2, 416,667 3
With VMware, assume all new hardware is purchased at the beginning of the implementation project, and therefore all hardware costs are absorbed in year 1:
CostsHardware ( After ) = (38)($10000) + (38)($30000) + (4)($45000) = $1, 434, 000
4.5 Power
Power consumption in the data center can be categorized into two main categories: • Computing Infrastructure (IT loads): Server hardware, network switches, SAN components, etc. • Network Critical Physical Infrastructure, or NCPI (non-IT loads): transformers, uninterruptible power supplies (UPS), power wiring, fans, air conditioners, pumps, humidifiers, and lighting. [4] A complete model would account for each category of power consumption provided above. For simplicity, however, this document focuses only on the power saved from reduction in server hardware. Given this frame of reference, the cost of power is calculated by estimating the difference in power consumption of server hardware before and after virtualization. The power consumed by server hardware can be calculated by adding up the power ratings of each server in the data center. Because this number represents the maximum power used, it should be de-rated to achieve steady-state power consumption. The steady-state constant was determined empirically. According to the American Power Conversion Corporation “…the nameplate rating of most IT devices is well in excess of the actual running load by a factor of at least 33%.” [4] Forrester Research, Inc. corroborates this idea, indicating that idle x86 servers consume between 30%-40% of maximum (rated) power. [10] Power consumption is often calculated based on the form factor of the server (1U, 2U, 4U, etc). The methodology assumes that CPU correlates to U’s according to the following details: • 1 CPU = 1U • 2 CPU = 2U • 4 CPU = 4U • 8 CPU = 6U • 16 CPU = 12U • 32 CPU = 24U Inputs
Input STTLPWR Description Sum of the nameplate power ratings of all computing Default Value Before/After 1 CPU: 475W / 550W 2 CPU: 550W / 675W 4 CPU: 950W / 1150W Source Available from manufacturer’s web site (server specs). Also possible to use capacity
11
VMware Infrastructure 3 TCO Methodology
infrastructure in the data center in kW. E$ λ Price per hour of 1 kW of electricity. Steady-state constant
8 CPU: 1600W / 1900W 16 CPU: 4400W / 5200W 32 CPU: 9200W / 11000W $.0813 (Average commercial value for United States in 2005) 0.67
tools such as Dell’s Data Center Capacity Planner.2 Energy Information Administration [2] American Power Conversion [4]. On average, nameplate ratings are 33% higher than steady state load
TCO Calculation
Savings Power = λE$ (S TTLPWR )Before − (S TTLPWR ) After
[
]
Assumptions • CPU utilization will increase with VMware, and therefore power consumption will also increase. This increase in power is accounted for by increasing the power consumption per machine before and after virtualization. The exact correlation between power consumption and CPU utilization is difficult to achieve, since power consumption by processor varies by model and brand. However, using the Dell online sizing tool, it is possible to estimate the change in power consumption based on load type. Dell offers the following load categories: Idle, Average Load, I/O Intensive, and Processor Intensive. A processor intensive load may consume up to 150W more than an idle load, and 80W more than an average load according to the tool. (Note: the tool does not provide actual utilization percentages for each of the categories given.)
Example Bartok, Inc. pays $0.0813 per kWh of electricity. The power ratings of Bartok, Inc. hardware are given in the following table:
Type 1 CPU 2 CPU 4 CPU 8 CPU Before Quantity 300 500 200 0 Power Rating 475W (0.475 kW) 550W (0.55 kW) 950W (0.95 kW) 1600W (1.60 kW) After Quantity 0 38 38 4 Power Rating 550W (0.550 kW) 675W (0.675 kW) 1150W (1.150 kW) 1900W (1.90kW)
The yearly server power costs before VMware are:
CostsPower ( Before ) = (0.67)($0.0813/ kWh) ×
[(300)(0.475kW ) + (500)(0.55kW ) + (200)(0.95kW )] ⎢1day ⎥ ⎢ 1year
The yearly server power costs after VMware are:
⎡ 24h ⎤ ⎡ 365day ⎤ ⎥ = $289,878 ⎣ ⎦⎣ ⎦
2
http://www.dell.com/html/us/products/rack_advisor_test/index.html 12
VMware Infrastructure 3 TCO Methodology
CostsPower ( After ) = (0.67)($0.0813 / kWh) ×
[(38)(0.675kW ) + (38)(1.15kW ) + (4)(1.90kW )] ⎢
⎡ 24h ⎤ ⎡ 365day ⎤ ⎥⎢ ⎥ = $36, 718 ⎣1day ⎦ ⎣ 1 year ⎦
4.6 Cooling
All data center electrical equipment produces heat. Heat producing items include IT equipment (servers, switches, SAN components), power delivery systems (UPS, Power Distribution), Air Conditioning Units, Lighting, and even people. [5] For simplicity, this section focuses solely on the heat produced by server hardware. Data center design plays an important role in determining the thermal efficiency and the cost of cooling. Many data centers still employ a front-to-back layout, which positions servers in the same direction. This means the heat emission from the back of one server feeds directly into the air intake of the front of another server. A better approach is the hot-aisle/cold-aisle layout (depicted below) which mitigates the unacceptable temperature gradients associated with front-to-back layouts. [10]
Even with an optimized data center layout, as much as 25% airflow redundancy is still required. [1] In the event of a CRAC (Computer Room Air Conditioning) unit failure, airflow redundancy will continue to satisfy cooling requirements. Furthermore, many data centers have hot spots, where heat density is greater than other areas. Focused redundancy satisfies localized data center cooling requirements. [11] Beyond requirements for airflow redundancy, data centers require additional airflow to account for inefficiencies related to humidification. Humidification is required to reduce the potential for damage resulting from a static discharge. [11] Most air conditioning systems, however, induce humidity loss that is caused by the air-cooling function of the air conditioning system initiating condensation of water vapor. To maintain an acceptable humidity level, supplemental humidification is required, which creates additional load on CRAC units. [5] The additional over-sizing can be as much as 30% of the standard load. [5]
13
VMware Infrastructure 3 TCO Methodology
As a first-order approximation, the AC power consumed in the data center is completely converted to heat. Therefore, the power rating in Watts of server hardware is equal to thermal output. [5] Furthermore, according to experiments completed in the HP Laboratories, 0.8W of power is consumed by the cooling equipment for every 1W of heat dissipation in the data center (designated in this document as the Load Factor, L). This figure is confirmed by Forrester Research, Inc., which estimates that 0.5W to 1.0W of power is required to dissipate 1W of heat. [10] Given the above information, the total cost of cooling can be calculated. Inputs
Input STTLPWR Description Sum of the nameplate power ratings of all computing infrastructure in the data center in kW. Price per hour of 1 kW of electricity. Steady-state constant Default Value Before/After 1 CPU: 475W / 550W 2 CPU: 550W / 675W 4 CPU: 950W / 1150W 8 CPU: 1600W / 1900W $.0813 (Average commercial value for United States in 2005) 0.67 Source Available from manufacturer’s web site (server specs). Also possible to use capacity tools such as Dell’s Data Center Capacity Planner.3 Energy Information Administration [2] American Power Conversion [4]. On average, nameplate ratings are 33% higher than steady state load Empirically determined in HP Laboratories. [3]
E$ λ
L
ρ
δ
Cooling Load Factor amount of power consumed by the cooling equipment for 1W of heat dissipated Airflow Redundancy Constant – redundant airflow required to cool data center Inefficiency (Humidification) Constant – redundant airflow to account for burden of humidification
0.8
25%
SearchDataCenter.com [1]
25%
SearchDataCenter.com [1]
TCO Calculation
Savings Cooling =
λLE $ (1 + ρ ) (S TTLPWR )Before − (S TTLPWR ) After δ
[
]
Assumptions • This calculation does not include the amortization and maintenance costs of the power delivery and cooling systems. For explicit calculation of this value, refer to “Cost Model for Planning, Development, and Operation of a Data Center.” [3] The value was excluded here to achieve simplicity in the model.
3
http://www.dell.com/html/us/products/rack_advisor_test/index.html 14
VMware Infrastructure 3 TCO Methodology
Example The power consumed by Bartok, Inc.’s servers is repeated below for reference.
Type 1 CPU 2 CPU 4 CPU 8 CPU Before Quantity 300 500 200 0 Power Rating 475W (0.475 kW) 550W (0.55 kW) 950W (0.95 kW) 1600W (1.60 kW) After Quantity 0 38 38 4 Power Rating 550W (0.550 kW) 675W (0.675 kW) 1150W (1.15 kW) 1900W (1.90kW)
Bartok, Inc. calculates the rated power of the servers by adding all of the nameplate ratings and determines that 407 kW of steady-state power is consumed before VMware and 51.6 kW are consumed after VMware is implemented. These figures also represent the amount of heat that is emitted in Bartok, Inc.’s data center. The cost to cool this heat emission in the current environment is therefore:
⎡ 24h ⎤ ⎡ 365day ⎤ CostCooling ( Before ) = [(0.8)($0.0813/ kWh)(1.25)(1.25)][ 407kW ] ⎢ ⎥⎢ ⎥ = $362,348 ⎣1day ⎦ ⎣ 1year ⎦
Using VMware, these costs are reduced to the following:
⎡ 24h ⎤ ⎡ 365day ⎤ CostCooling ( After ) = [(0.80)($0.0813/ kWh)(1.25)(1.25)][51.6kW ] ⎢ ⎥⎢ ⎥ = $45,897 ⎣1day ⎦ ⎣ 1year ⎦
4.7 SAN Environment
The SAN architecture is likely to change significantly using VMware Infrastructure 3. Virtualization is a significant driver for moving to a shared storage infrastructure, and companies implementing VMware Infrastructure 3 are likely to invest in more shared storage. However, companies often are convinced of the benefits of shared storage even before implementing virtualization, but costs associated with implementing shared storage appear prohibitive. Server consolidation with VMware Infrastructure 3 dramatically reduces these costs by the reducing the number of SAN switches and HBAs. The reduction in switches and HBAs resulting from virtualization yields the change in total cost of ownership. A diagram of a typical SAN Architecture is depicted below. In the diagram, the “SAN Fabric” houses the switches that represent a significant portion of the reduction in TCO.
15
VMware Infrastructure 3 TCO Methodology
Inputs
Input NSANSW Description Number of new SAN switches per year Number of new HBAs per year Price per HBA Price per SAN switch Price per terabyte of storage SAN Storage Default Value Calculated from total number of servers (assume switches are fully redundant) 2 HBA’s per server that is refreshed $1,000 $5,000 $6,000 Before: customer data After: Assume 20GB per virtual machine (2 vmdk files at 10GB per file) 3 Years 25% before, 100% after (all VM’s will be on the SAN) Source Customer data
NHBA P$(HBA) P$(SWITCH) P$(STORAGE) σTOTAL
Customer data Survey of several HBA’s (Fibre Channel) from CDW4 Survey of several switches from CDW Survey of storage from CDW 20GB per VM is a conservative estimate received from a customer. Some customers report 4GB per VM. Industry average. VMware estimate. Industry data – source not available
SLIFE n
Useful life of server (also called the refresh rate) Percent of servers attached to SAN
TCO Calculation
)⎤ )⎤ ⎡ ( N )( P ⎡ ( N )( P SavingsHBA = ⎢ HBA $( HBA) ⎥ − ⎢ HBA $( HBA) ⎥ S LIFE S LIFE ⎣ ⎦ Before ⎣ ⎦ After
4
http://www.cdw.com 16
VMware Infrastructure 3 TCO Methodology
)( P )⎤ )( P )⎤ ⎡ (N ⎡ (N SavingsSANSW = ⎢ SANSW $( SANSW ) ⎥ − ⎢ SANSW $( SANSW ) ⎥ S LIFE S LIFE ⎣ ⎦ Before ⎣ ⎦ After
⎡σ P ⎤ ⎡σ P ⎤ SavingsSTORAGE = ⎢ $( STORAGE ) ⎥ − ⎢ $( STORAGE ) ⎥ ⎣ S LIFE ⎦ Before ⎣ S LIFE ⎦ After
Because we assume 2 HBA’s per server, the number of HBA’s is calculated as follows:
N HBA = 2n( STOTAL )
Similarly, the number of SAN Switches is calculated as follows. The calculation is multiplied by 2 to account for redundant switches:
N SANSW = 2n( STOTAL / 24)
Note: this formula behaves as a step function, and STOTAL / 24 should always be rounded up to the nearest whole number. For example, if there are only 6 servers in the environment, STOTAL / 24 = .25 . To achieve a whole number value of switches, we round to 1. Therefore, the total number of switches required is 2*1 = 2.
Assumptions • There are other storage savings, such as reduced cabling, less power consumed, etc. However, for simplicity reasons, these items are omitted. • SAN Switches, HBA’s, and SAN storage are refreshed at the same rate as server hardware • There are 2 HBA’s per server • SAN switches are fully redundant • For modeling purposes, a factor can be included to indicate the percentage of servers attached to the SAN. The calculations here assume all servers that are virtualized are attached. Example In the current Bartok, Inc. example, Bartok is moving from 1000 to 80 physical servers. The customer negotiates the following prices for SAN components:
P$( HBA) = $1,000 P$( SANSW ) = $5,000 P$( STORAGE ) = $6,000
First, calculate the number of networking and storage units required. Since 1000 workloads will be virtualized, SAN storage will increase according to the following calculation:
17
VMware Infrastructure 3 TCO Methodology
⎛ 20GB ⎞⎛ 1TB ⎞ Storage = 5TB + (1000Workloads) ⎜ ⎟⎜ ⎟ = 25TB ⎝ Workloads ⎠⎝ 1000GB ⎠
Assuming 2 HBA’s per server, 25% of servers connected to the SAN before, and 100% of servers connected to the SAN after, the total number of HBAs required are:
N HBA( Before ) = 2(1000)(25%) = 500 N HBA( After ) = 2(80)(100%) = 160
Assuming redundant switches, the total switches required are:
N SANSW ( Before ) = 2[(1000)(25%) / 24] = (2)(10.42) = (2)(11) = 22 N SANSW ( After ) = 2[(80)(100%) / 24) = 2(3.33) = 2(4) = 8
The network and storage savings (yearly) are now calculated as follows: HBAs:
Cost HBA( Before ) = Cost HBA( After )
(500)($1000) = $166,667 3 (160)($1000) = = $53,333 3
SAN Switches:
CostSANSW ( Before ) = CostSANSW ( After )
Storage:
(22)($5000) = $36,667 3 (8)($5000) = = $13,333 3
CostStorage ( Before ) =
CostStorage ( After )
(5)($6000) = $10,000 3 (25)($6000) = = $50,000 3
4.8 Network
The number of physical network components is dramatically reduced with Virtual Infrastructure. With fewer physical servers attached to the network, fewer switches, NIC’s and cables are required to provide server connectivity. The diagrams below illustrate this idea.
18
VMware Infrastructure 3 TCO Methodology
Virtualized Environment
Typically, the number of NIC’s per server is increase in a virtualized scenario (3 per server), although the total number of NIC’s across all servers is substantially reduced. For the TCO calculation, the reduced NIC requirements are accounted for in the price of the server hardware, since they often accompany server purchase. Therefore, this section concentrates only on the reduction in network switches. The savings in cabling are small and therefore are not recorded here. Inputs
Input NNETSW Description Number of new network switches per year Default Value Calculated from total number of servers Source Customer data
19
VMware Infrastructure 3 TCO Methodology
SNIC pn P$(NETSW)
Number of NIC’s per server Number of ports per NIC Price per SAN switch
Before: 2 per server After: 3 per server 1 $4,000
Surveyed customers Customer data – highly dependent on types of NIC’s being used. Surveyed customers and various switches at CDW
TCO Calculation
)( P )⎤ )( P )⎤ ⎡ (N ⎡ (N SavingsNETSW = ⎢ NETSW $( NETSW ) ⎥ − ⎢ NETSW $( NETSW ) ⎥ S LIFE S LIFE ⎣ ⎦ Before ⎣ ⎦ After
NNETSW is calculated as follows
N NETSW = 2( S NIC )( pn )( STOTAL ) / 24
where the denominator represents the number of ports per network switch. Note: this formula behaves as a step function, and STOTAL / 24 should always be rounded up to the nearest whole number.
Assumptions • Other savings resulting from reduced cabling and less power consumption by network switches are omitted for simplicity. • The costs of NIC’s are accounted for in the price of the servers, and are therefore not shown here. • Network switches have the same refresh rate as servers (3 years)
Example Bartok, Inc’s requirement for network switches is expected to decrease significantly after moving from 1000 to 80 physical servers. In its current data center, Bartok has on average 2 NIC’s per server and each NIC has one port. In the future state, Bartok is expecting to increase to 3 NIC’s per server, with each NIC still having one port. Furthermore, each network switch has 24 ports that are used, and costs $4,000. First, calculate the number of total switches required:
N NETSW ( Before ) = (1000)(2)(1) / 24 = 83.33 = 84 N NETSW ( After ) = (80)(3)(1) / 24 = 10
The yearly cost of switches is now calculated as follows:
Cost NETSW ( Before ) =
(84)($4000) = $112,000 3
20
VMware Infrastructure 3 TCO Methodology
Cost NETSW ( After ) =
(10)($4000) = $13,333 3
4.9 Data Center Real Estate
Savings in data center real estate are achieved by examining the potential reduction in required space after VMware Infrastructure 3 is implemented. Due to the special infrastructure (cooling, power systems, etc) required in data centers, they are often significantly more expensive to build than standard commercial properties. A data center rated at 40W per square foot costs approximately $400 per square foot. [8] At the 2009 projection of 500W per square foot, the same data center would cost $5000 per square foot to build. [8] VMware can reduce a company’s data center footprint, and prevent future construction of other data centers. The TCO calculation accounts for the yearly data center carrying costs, but does not examine the prevention of construction of other data centers. Inputs
Input A$ Description Cost per square foot to build data center Default Value $600 per square foot Source Based on size of 60W per square foot (conservative) [9] and $400/square foot for 40W/square foot data center [8] (Note: in 2005, average size was 80W/square foot) Dell PowerEdge Rack 42105 (23.94” x 39.83”) Customer approximation Calculation Depends on customer. 6% used as an estimate.
RSF RSPACE RTOTAL r
Square feet per rack Percent of space used by racks Total number of racks Annual Percentage Rate
7 Square Feet (approx) 30% Supplied by customer or calculated. See below. 6.00%
TCO Calculation
SavingsREALESTATE = [ ( MonthlyCost )(12) ]Before − [ ( MonthlyCost )(12)] After
Assumptions • The additional space required for SAN’s is small and not included here. For example, an EMC CLARiiON CX500 can hold up to 4 TB in a 4U configuration. In the example here, the extra SAN space would only occupy 1 24U rack. Example Bartok, Inc. loads 24U per rack (recall that for simplicity U’s are roughly correlated to CPU’s). Each rack requires 7 square feet of floor space, and racks take up 30% of the total floor space in the data center. The cost of the data center is approximately $600 per square foot, based on the 60W per square foot of power that is required.
5
http://www.dell.com/downloads/global/products/pedge/en/rack_system.pdf 21
VMware Infrastructure 3 TCO Methodology
First, determine the number of racks that are required before and after virtualization
RTOTAL ( Before ) = RTOTAL ( After )
(300)(1) + (500)(2) + (200)(4) = 88 Racks 24 (38)(2) + (38)(4) + (4)(8) = = 10 Racks 24
The space required for the computing infrastructure is now calculated as follows:
Re quiredSpace( General ) = Re quiredSpace( Before ) = Re quiredSpace( After ) =
( RTOTAL )( RSF ) RSPACE
(88)(7) = 2,053 ft 2 0.30
(11)(7) = 257 ft 2 0.30
The total cost of the data center before and after virtualization can now be calculated:
Cost REALESTATE ( Before ) = (2,053)($600) = $1, 232,000 Cost REALESTATE ( Before ) = (257)($600) = $154,000
To determine the monthly (and yearly) carrying costs, the total cost of the data center can be amortized over the life of the data center (assume 10 years). Bartok’s annual percentage rate on the cost of the data center is r = 6.00%, making the periodic (monthly) rate 6.00/12 or .500%. The annuity factor (the present value of $1 paid for t=120 periods) is calculated as follows:
AF =
⎞ 1 ⎛ 1 = 90.07 ⎜1 − (10)(12) ⎟ .005 ⎝ (1 + .005) ⎠
Given this annuity factor, the monthly payments are:
PaymentMonthly ( Before ) = PaymentMonthly ( Before ) =
$1, 232,000 = $13,678 90.7 $154,000 = $1,710 90.7
Given these monthly payments, the yearly costs for the data center before and after virtualization are:
22
VMware Infrastructure 3 TCO Methodology
CostYearly ( Before ) = ($13,678)(12) = $164,133 CostYearly ( After ) = ($1,710)(12) = $20,517
23
VMware Infrastructure 3 TCO Methodology
5 Business Continuity
3 offers significant savings by improving Business Continuity. These savings can be categorized into several areas: • Disaster Recovery CapEx and OpEx – The savings achieved in a company’s primary data center based on server consolidation using VMware can also be achieved in the DR site. • Disaster Recovery Maintenance and Provisioning – Maintenance and provisioning that occurs in the DR site as a result of server refreshes can be made quicker and easier. • General Downtime – VMware can dramatically reduce planned downtime associated with scheduled software upgrades, application maintenance, hardware reconfiguration, etc. In addition, it reduces unplanned downtime such as unscheduled outages due to hardware failure, application (software) failure, etc. • Recovery from a massive failure: Unsuccessful or lengthy recoveries are commonplace, and VMware Infrastructure 3 can speed recovery and enable long-term survival In addition to the items listed above, further savings can be achieved through simplified disaster recovery planning and testing. These savings are mentioned here but not quantified below.
5.1 Disaster Recovery CapEx and OpEx
Implementing VMware at the Disaster Recovery site can yield similar reductions in hardware and operations expenses that are achieved in a customer’s primary data center. These reductions are calculated in the same manner as outlined in Section 4 of this document, and are therefore omitted here.
5.2 Disaster Recovery Maintenance and Provisioning
Servers in the Disaster Recovery site require maintenance as well as provisioning. The maintenance is not quantified here, but should be mentioned as a soft cost in TCO analysis. Savings resulting from workload provisioning should be calculated using the methodology given in Section 6 of this document.
5.3 Downtime
VMware Infrastructure 3 can help reduced both planned and unplanned downtime. Planned hardware maintenance can result in a temporary disruption of business, since servers have to be powered down. VMware’s VMotion technology drastically reduces this downtime, though, by allowing IT administrators to move running virtual machines between servers. The cost benefits of planned downtime reduction are not quantified here, but should be mentioned in TCO discussions as a soft cost. Unplanned downtime is also reduced using VMware Infrastructure 3. VMotion allows administrators to proactively move virtual machines away from overloaded hosts or from servers that have experienced hardware failure. This minimizes interruption to normal business activity, and can ultimately lead to preservation of revenue. Often customers understand the cost of downtime to their business, as well as their average downtime on a yearly basis. Using these figures, it is possible to estimate the savings resulting from reduced downtime after implementing VMware. Again, only unplanned downtime is quantified below.
24
VMware Infrastructure 3 TCO Methodology
Inputs
Input D$ τU kU Description Cost of downtime per hour Unplanned downtime per year Unplanned downtime reduction Default Value Customer dependent 15 hours 75% Source Customer input Customer input VMware projection (conservative)
TCO Calculation
SavingsDowntime = D$τ U (1 − kU )
Assumptions • The unplanned downtime reduction factor is currently a conservative estimate based on discussions with VMware engineers and sales people. Empirical data should be added when made available.
Example Bartok, Inc. knows its average downtime and cost of downtime to have the following characteristics:
Category Cost of downtime (D$) Unplanned downtime per year (τU) Value $20,000 per hour 15 hours
Furthermore, they project that very conservatively VMware infrastructure 3 will allow them to cut unplanned downtime by 25%. The cost of unplanned downtime is calculated as follows:*
CostsUnplannedDowntime ( Before ) = ($20,000)(15) = $300,000 CostsUnplannedDowntime ( After ) = ($20,000)(3.75) = $75,000
* These calculations are the unfactored formulas that yielded the “Savings” formula in the TCO Calculation section above. Note that either method of cost calculation arrives at the same savings:
SavingsDowntime = ($20,000)(15)(1 − 0.25) = $225,000
5.4 Recovery from a Disaster or Massive Failure
In the event of a massive failure (disaster), VMware Infrastructure 3 can dramatically reduce recovery times. Traditional disaster recovery solutions require that recovery hardware is a
25
VMware Infrastructure 3 TCO Methodology
duplicate of production hardware. Server configuration and complex multi-step processes in these bare-metal recoveries are time-consuming and difficult. Hardware independence and single-step file recovery offered by VMware Infrastructure 3 enable improved recovery times and an expeditious return to business normalcy. The actual gains achieved using VMware Infrastructure 3 depend on the current disaster recovery infrastructure. For example, VMware Infrastructure 3 provides substantial benefits for customers who use array-based replication, and even greater benefits to customers who execute backups from tape. (Although there are other disaster recovery solutions, only these two are mentioned here for demonstration purposes). Input
Input D$ τR Description Cost of downtime per hour Time for full recovery after disaster Recovery reduction factor (% of original recovery time with virtualization) Default Value $20,000 40 hours (Placeholder for model – this is typically customer dependent) 25% Source Emerson6 Customer input
kR
VMware projection
TCO Calculation
SavingsRecov ery = D$τ R (1 − k R )
Assumptions • The customer should know their average recovery time from a disaster based on disaster recovery tests. If a customer does not perform disaster recovery testing, savings can still be projected using this methodology. • The cost of downtime includes all recovery costs in the event of a disaster (forgone revenue, labor costs, etc). • The recovery reduction factor is currently a conservative estimate based on discussions with VMware engineers and sales people. Empirical data should be added when made available. Example Through performing disaster recovery tests, Bartok, Inc discovers that it will take a minimum of 40 hours to fully recover in the event of a massive failure. Furthermore, the tests have shown that they will lose $20,000/hour of downtime during a failure. The ease of workload deployment offered by VMware Infrastructure 3 is expected to conservatively reduce their recovery times to 25% of the current time.
CostRecov ery ( Before ) = ($20,000)(40) = $800,000 CostRecov ery ( After ) = ($20,000)(10) = $200,000
6
http://www.continuitycentral.com/news02575.htm 26
VMware Infrastructure 3 TCO Methodology
6 Infrastructure Provisioning
Infrastructure provisioning involves two cost categories. The first is the cost of installing and configuring VMware Infrastructure 3 by customer administrators. The second is the cost reduction resulting from decreased provisioning time once VMware Infrastructure 3 is implemented.
6.1 Virtual Infrastructure Installations
Description Installing and configuring VMware Infrastructure 3 in the data center will require customer administrator time. This is different that the investment made in VMware Services, described in the “VMware Services”, which accounts for the labor of VMware service providers. It is expected that customer personnel will be participating in VMware implementation as well. Multiplying the average installation time by the number of installations yields the increase in labor hours. Inputs
Input IESX tESX Description Number of ESX Installations Hours per ESX Installation Default Value Number of servers after virtualization 2 Source Derived from customer input of total number of servers Survey of customers
TCO Calculation
⎛I t ⎞ ⎛I t ⎞ HoursESX INSTALL = ⎜ ESX ESX ⎟ − ⎜ ESX ESX ⎟ ⎝ S LIFE ⎠ Before ⎝ S LIFE ⎠ After
Assumptions • The above calculation assumes that customers are familiar with VMware installation. • Most customers will measure the savings (or further investment) in labor as a function of hours rather than dollars. That is because it is unlikely that the administrator headcount will be augmented as a result of virtualization. Instead, they will be deployed to other activities as necessary. Example Bartok, Inc. is ready to install ESX on the 80 servers it has recently purchased as part of the virtualization project. System administrators have gone through training and are prepared to do the installations. The hours spent on installations are calculated as follows:
HoursESX INSTALL ( Before ) = (80)(2) = 160hours
Assuming Bartok, Inc. pays $60/hour for System administrators, this equates to the following investment:
CostsESX INSTALL = (160)($60) = $9,600
27
VMware Infrastructure 3 TCO Methodology
6.2 Provisioning
Infrastructure provisioning is made significantly easier through using VMware Infrastructure 3, since it allows administrators to provision workloads from their desk without having to acquire and implement new hardware. Provisioning savings are most reliably measured in hours. The many labor hours saved will result in redeployment of administrators to value added activities. However, it is possible to associate a dollar value to provisioning savings by multiplying the hours saved by an administrator hourly wage. Inputs
Input WPROV tPROV Description Total workloads provisioned per year Total hours to provision a workload Default Value NA 20 hours (without VMware) 1 hour (with VMware) Source Derived from total servers refreshed per year Typically 20 hours without VMware, and 1 hour with VMware. This figure is an estimate based on review of customer case studies and internal VMware projections. However, the provisioning time reduction factor can also be used (see next Input). Based on a sample of 21 customers [6]
kPT
Provisioning time reduction factor – factor by which provisioning time is reduced
13.48 hours
TCO Calculation Method 1
SavingsPROVISIONING = (WPROV t PROV ) Before − (WPROV t PROV ) After
Method 2
SavingsPROVISIONING = (WPROV t PROV ) Before −
(WPROV tPROV ) After
k PT
Assumptions • The customer may not actually see the bottom line altered as a result of provisioning savings. This is because resources are likely to be deployed on other tasks, such as disaster recovery, if not spending time on provisioning servers. This still represents “savings”, however, as more value-added activities are addressed as the result of virtualization. Example 1 Recall that Bartok, Inc. currently has 1000 physical servers (workloads) and refreshes hardware every three years (333 workloads per year). Bartok’s average provisioning time is
28
VMware Infrastructure 3 TCO Methodology
20 hours. After implementing VMware, Bartok expects its provisioning time to be reduced to 1 hour per workload. This information is illustrated in the following table:
WPROV tPROV
Before VMware 333 20
After VMware 333 1
The total hours to provision workloads before and after VMware are calculated as follows:
HoursPROVISIONING ( Before ) = (333)(20) = 6660 hours HoursPROVISIONING ( After ) = (333)(1) = 333 hours
Example 2 Instead of projecting a new provisioning time of 1 hour as offered in Example 1, Bartok uses the provisioning time reduction factor of 13.48. The hours before and after installation are now:
HoursPROVISIONING ( Before ) = (333)(20) = 6660 hours HoursPROVISIONING ( After ) =
(333)(20) = 494 hours 13.48
29
VMware Infrastructure 3 TCO Methodology
7 Summary
This section provides the TCO summary of Bartok’s implementation of VMware.
7.1 Net Present Value
The calculations in the above sections yield future savings on a yearly basis. To determine the final TCO reduction, all dollar amounts are converted to present values. The model and methodology assume that savings are realized after the implementation of VMware Infrastructure 3 is complete. The length of the implementation is an input in the model, and changing it alters the net present value of the savings. The default value for implementation is 6 months, which implies that the first discount period is 1.5 years, the second is 2.5, and the third is 3.5. The diagram below illustrates this concept.
t=0
t=0.5
t=1.5
t=2.5
t=3.5
Year 1 Implementation (Year 0)
Year 2
Year 3
NPV calculations are not shown here. This document assumes the reader is familiar with discounting cash flows.
7.2 Bartok’s 3 Year TCO Reduction
The calculations in the document have been aggregated here to show Bartok’s overall TCO reduction over a 3 year period. The image below was taken from the TCO Calculator that is derived from the methodology of this document.
0 VMware Investment VMware Software VMware Services VMware Training $ (747,500) $ (414,078) $ (129,300) $ (1,290,878) $ $ $ $ Yearly Savings 1 2 (186,875) (186,875) $ $ $ $ (186,875) (186,875) $ $ $ $ 3 (186,875) (186,875) $ $ $ $ Total (1,308,125) (414,078) (129,300) (1,851,503)
Monetary Savings (Cash Flows) Server Hardware Storage Network Power and Cooling Data Center Real Estate Disaster Recovery Downtime Provisioning
$ (1,434,000) $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ $ (1,434,000) $
2,416,667 96,667 98,667 569,611 143,616 600,000 225,000 360,726 4,510,953
$ $ $ $ $ $ $ $ $
2,416,667 96,667 98,667 569,611 143,616 225,000 370,326 3,920,553
$ $ $ $ $ $ $ $ $
2,416,667 96,667 98,667 569,611 143,616 225,000 370,326 3,920,553
$ $ $ $ $ $ $ $ $
5,816,000 290,000 296,000 1,708,833 430,848 600,000 675,000 1,101,379 10,918,060
Total Yearly Cash Flow Discount Period Net Present Value
$ (2,724,878) $ 0 $ (2,724,878) $
4,324,078 1.00 3,793,051
$ $
3,733,678 2.00 2,872,944
$ $
3,733,678 3.00 2,520,127
$ $
9,066,557 6,461,244
30
VMware Infrastructure 3 TCO Methodology
The above calculations leveraged a 14% discount rate. Using this value, the following financial metrics were achieved: Financial Performance NPV IRR ROI Payback Period $6,461,244 140% 375% 8.32 Months
The diagram below provides a graphical representation of present value TCO reduction.
Present Value TCO Comparison over 3 Year Horizon
$12,000,000 Network VMware Training $8,000,000 Costs Provisioning Downtime Disaster Recovery $6,000,000 Data Center Real Estate Power and Cooling $4,000,000 VMware Services VMware Software Storage $2,000,000 Server Hardware
$10,000,000
$PV: Without VMware PV: With VMware
31
VMware Infrastructure 3 TCO Methodology
Appendix A: References
[1] McFarlane, Robert, “Let’s Add an Air Conditioner,” SearchDataCenter news article, published November 30, 2005. http://searchdatacenter.techtarget.com/columnItem/0,294698,sid80_gci1148906,00.html [2] Energy Information Administration, “Average Retail Price of Electricity to Ultimate Customers by End-Use Sector, by State.” http://www.eia.doe.gov/cneaf/electricity/epm/table5_6_b.html [3] Patel, Chandrakant D., Shah, Amip J., “Cost Model for Planning, Development, and Operation of a Data Center,” Internet Systems and Storage Laboratory, HP Laboratories, Palo Alto, June 9, 2005 [4] Sawyer, Richard, “Calculating Total Power Requirements for Data Centers”, American Power Conversion, 2004 [5] Rasmussen, Neil, “Calculating Total Cooling Requirements for Data Centers”, American Power Conversion, 2003 [6] “CustomerCaseStudies - OpExSavings.xls” (INTERNAL USE ONLY) [7] Disaster Recovery. (2006, July 13). In Wikipedia, The Free Encyclopedia. Retrieved July 17, 2006, from http://en.wikipedia.org/wiki/Disaster_recovery [8] Anthes, Gary, “Data Centers Get a Makeover”, Computerworld news article, published November 1, 2005. http://www.computerworld.com/databasetopics/data/datacenter/story/0,10801,97021,00.html?S KC=home97021 [9] Hughes, Ron, “The data center of the future – Part 1 – Current trends”, The Data Center Journal news article, published May17, 2005. http://www.datacenterjournal.com/News/Article.asp?article_id=315 [10] Fichera, Richard, “Power And Cooling Heat Up The Data Center”, Forrester Research, Inc. March 8, 2006 [11] Dunlap, Kevin, and Rasmussen, Neil, “The Advantages of Row and Rack-Oriented Cooling Architectures for Data Centers”, American Power Conversion, TCO
32
VMware Infrastructure 3 TCO Methodology
Appendix B: Conversions
12,000 BTU/h = 1 ton (http://en.wikipedia.org/wiki/BTU) 12,000 BTU/h = 3516 watts (http://en.wikipedia.org/wiki/Ton) 1 kilowatt = 0.283 tons of AC
33
VMware Infrastructure 3 TCO Methodology
Appendix C: Links
http://searchwinsystems.techtarget.com/originalContent/0,289142,sid68_gci1046212,00.html http://news.zdnet.com/2100-9584_22-6067896.html http://searchdatacenter.techtarget.com/columnItem/0,294698,sid80_gci1148906,00.html http://expertanswercenter.techtarget.com/eac/blog/0,295203,sid63_tax302039,00.html http://sanjose.bizjournals.com/sanjose/stories/2006/02/27/story5.html http://www.computerworld.com/databasetopics/data/datacenter/story/0,10801,97021,00.html?S KC=news97021 http://www.itjungle.com/tfh/tfh020705-story03.html http://www.dntp.com/news/pdfs/Design%20Guidelines%20for%20High%20Density%20Data%2 0Centers.pdf http://www.upsite.com/TUIpages/whitepapers/tuiheat1.0.html http://www.archinetix.com/heat-density-costs/
34
VMware Infrastructure 3 TCO Methodology
Appendix D: Pricing Regressions
The cost of services varies by the number of servers being virtualized. This cost is typically estimated on a per project basis, however, some general data points have been provided by Professional Services. The data points and corresponding regressions are provided below.
VMware Services Planning and Design # of Servers 200 $ 500 $ 1000 $ Slope 92607 Cost 100,000 175,000 250,000 Intercept -393627
# of Servers Deployment 50 $ 100 $ Slope 1100
Cost 125,000 180,000 Intercept 70000
Planning and Design Cost vs. Number of Servers $300,000 $250,000 $200,000 Cost $150,000 $100,000 $50,000 $0 200 400 600 800 1000 1200 y = 1100x + 70000 R2 = 1 Deployment Log. (Planning and Design) Linear (Deployment) y = 92607Ln(x) - 393627 R 2 = 0.9936 Planning and Design
Number of Servers
35
VMware Infrastructure 3 TCO Methodology
36
