it optimization to meet business goals. · 2008-11-07 · it optimization to meet business goals....
TRANSCRIPT
IT optimization to meet business goals.
Page 2
2 Introduction
3 Running IT like a business:
what does it mean?
4 Understanding the trends and
challenges of IT optimization
5 Choosing a model that
matches the business
7 Creating an IT strategy designed
to support business goals
10 Making connections between
IT and business
11 Conclusion
ContentsIntroduction
IT optimization to meet business goals.
Page 3
Highlights
Running IT like a business: what does it mean?
There is a delicate balance between
efficiency and innovation, cost
and quality.
By reducing infrastructure costs
and reinvesting the savings, CIOs
can drive both cost reduction
and business enablement and
innovation.
CIOs today are expected to help
drive business innovation and
ignite the integration of business
and technology.
IT optimization to meet business goals.
Page 4
HighlightsUnderstanding the trends and challenges of IT optimization
IT is expected to simplify the
existing infrastructure while
creating a flexible services portfolio
that can quickly adapt to changing
business conditions.
IT optimization to meet business goals.
Page 5
Highlights
Choosing a model that matches the business
An organization must first determine
the needs and characteristics that
drive its use of technology.
IT optimization to meet business goals.
Page 6
Highlights
Commodity organization
Utility organization
Partner organization
IT must leverage its resources
in order to build or maintain
technologies that support the
requirements of each customer
segment.
There are four customer segment
profiles: commodity, utility, partner
and enabler.
IT optimization to meet business goals.
Page 7
HighlightsEnabler organization
Creating an IT strategy designed to support business goals
The next step is to create an IT
optimization strategy that supports
the organization’s customer
segment profiles–and, therefore,
its business goals.
IT optimization to meet business goals.
Page 8
HighlightsDefining the components of the IT infrastructure
Management
Seven domains typically constitute
the IT infrastructure for almost any
organization.
Each domain falls into one of two
groups—management or technical.
IT optimization to meet business goals.
Page 9
Highlights
Technical
Management domains are strategic
alignment, processes, organization,
and finance and environment.
Technical domains are
infrastructure, network, and
applications and data.
IT optimization to meet business goals.
Page 10
HighlightsMaking connections between IT and business
Collectively, the domains define
the potential areas of analysis
and applied expertise needed to
achieve IT optimization in today’s
competitive landscape.
Enable an adaptable IT infrastructure
Strategicalignment Processes Organization
Finance andenvironment Infrastructure
Applications and dataNetwork
Management domains Technical domains
Improve internal efficiencies
Reduce overall IT cost
Promote customer satisfaction
Enable new business capabilities
Provide consistent delivery service
Primary focus
Streamline technicalarchitecture/vendors
Reduce total IT spending
Establish global consistencyof IT management
Optimize use of assets
Maintain data more effectively
Secondary focus
IT optimization to meet business goals.
Page 11
HighlightsConclusion
For more information
ibm.com/services
To innovate and implement the
best optimization strategy, try
this process of performing a self-
evaluation, creating a strategy
and aligning IT with business.
© Copyright IBM Corporation 2007
IBM Global Services
Route 100
Somers, NY 10589
U.S.A.
Produced in the United States of America
01-07
All Rights Reserved
IBM and the IBM logo are trademarks or registered
trademarks of International Business Machines
Corporation in the United States, other countries,
or both.
IT Infrastructure Library is a registered trademark
of the Central Computer and Telecommunications
Agency which is now part of the Office of Govern-
ment Commerce.
ITIL is a registered trademark, and a registered
community trademark of the Office of Government
Commerce, and is registered in the U.S. Patent
and Trademark Office.
Other company, product and service names may
be trademarks or service marks of others.
IBM assumes no responsibility regarding the
accuracy of the information provided herein
and use of such information is at the recipient’s
own risk. Information herein may be changed
or updated without notice. IBM may also make
improvements and/or changes in the products
and/or the programs described herein at any time
without notice.
References in this publication to IBM products
and services do not imply that IBM intends to
make them available in all countries in which
IBM operates.
1 “IBM Global CEO Study 2006.” IBM, 2006.
2 Prewitt, Edward and Ware, Lorraine Cosgrove.
“The State of the CIO ’06: The Survey.” CIO
Research, January 1, 2006; www.cio.com/state.
3 IBM and industry studies, client interviews.
4 IBM and industry studies, client interviews.
GSW00427-USEN-02
Driving business value with a
virtualized infrastructure.
Page 2
2 Introduction: expanding a
technical focus to address
business challenges
3 A growing popularity based
on virtualization’s ability to
optimize IT
4 Meeting boardroom
opportunities by improving
enterprise operations
7 Addressing business
issues that are key to the
organization’s growth
11 Providing a holistic view of
the business as well as its
technology
12 Building on a solid foundation
of intellectual capital and
expert assistance
ContentsIntroduction: expanding a technical focus to address business challenges
Driving business value with a
virtualized infrastructure.
Page 3
A growing popularity based on virtualization’s ability to optimize IT
Virtualization’s popularity is growing
rapidly, but many organizations
are using it primarily for technical
functions.
Highlights
Virtualization has widespread appeal based on benefits that range from reduced IT costs to simplified
IT environments, streamlined IT management and increased IT flexibility.3
Reduce costs
Simplify IT infrastructure and administration
Increase server utilization
Increase scalability of infrastructure
Enhance resilience and reliability
Improve flexibility to business goals/cycle
Improve application performance
Automate IT operations
Accelerate application development and deployment
Have a single view of the IT environment
Manage a heterogeneous server environment
Manage a heterogeneous storage environment
What applications on what servers. How they relate
Enable an SOA
Other
Virtualization motivators
48%
48%
29%
25%
16%
15%
11%
11%
10%
9%
6%
5%
4%
1%
57%
Source: IBM Systems and Technology Group (1Q06)
Driving business value with a
virtualized infrastructure.
Page 4
Highlights
Meeting boardroom opportunities by improving enterprise operations
Technical functions represent
only a portion of virtualization’s
potential benefits.
Virtualization can address strate-
gic business issues including
flexibility, responsiveness and the
ability to achieve innovation.
Driving business value with a
virtualized infrastructure.
Page 5
Highlights
Improving operations: power and cooling
If the boardroom opportunity is …
to improve operations and business processes as a means of achieving efficiencies …
CIOs must …
apply new technologies that can increase operational excellence and reduce complexity …
in order to …
reduce cost and enhance system availability, performance and flexibility.
Virtualization presents a significant
opportunity for the CIO to contribute
more to achieving business goals.
Driving business value with a
virtualized infrastructure.
Page 6
Highlights
Improving operations: dissimilar applications
Virtualization can be a powerful
solution for keeping power and
cooling usage—and costs—
in check.
The ability to consolidate dissimilar
applications can dramatically
reduce the need for physical
machines in the data center.
Driving business value with a
virtualized infrastructure.
Page 7
Highlights
Addressing business issues that are key to the organization’s growth
To be most effective, virtualization
must address concerns that tradi-
tionally have been the province
of business.
Driving business value with a
virtualized infrastructure.
Page 8
HighlightsEnsuring trusted data for insightful decisions
If the boardroom opportunity is …
to quickly respond to marketplace opportunities by being more customer focused …
CIOs must …
ensure that the right customer information can be delivered to the right people at the right time
in the right context …
in order to …
support faster, more informed business decisions.
The explosive growth of information
makes it critical to connect the
organization with an enterprisewide
view of data, not a siloed one.
Virtualization can improve the
sharing, use and management of
information—so business can
derive more value from its data.
Driving business value with a
virtualized infrastructure.
Page 9
Highlights
Aligning technologies, objectives and policies
If the boardroom opportunity is …
to create a strategic advantage by building an organization that is more fully integrated in
its operations …
CIOs must …
establish governance policies that align IT practices with business goals and procedures …
in order to …
break down management and operational silos, protect the business by better responding
to threats and risks, and comply with governmental and industry regulations.
Driving business value with a
virtualized infrastructure.
Page 10
Highlights
Instead of “future proofing”
environments with extra equipment,
organizations can adjust and
provision resources as needed.
Storage equipment can be tiered,
and service level agreements can
make assets available when and
where they are needed.
Virtualization can make data
gathered for compliance usable
for daily operations.
Driving business value with a
virtualized infrastructure.
Page 11
Highlights
Using flexibility to capture opportunity
If the boardroom opportunity is …
to identify new revenue streams that will help the organization grow profitably …
CIOs must …
provide capabilities that enable rapid business process change …
in order to …
respond to ever-changing customer and stakeholder needs.
Virtualization can meet customer
and stakeholder needs for higher
levels of business flexibility.
Driving business value with a
virtualized infrastructure.
Page 12
Highlights
Providing a holistic view of the business as well as its technology
Holistic management tools are key
to bridging technologies, breaking
down silos and reducing complexity.
Virtualization can provide the ideal
foundation for a flexible, service-
oriented architecture, and both
have the ability to add, subtract
and rearrange resources.
Driving business value with a
virtualized infrastructure.
Page 13
Highlights
Building on a solid foundation of intellectual capital and expert assistance
A holistically managed environment
help can transform IT from a service
provider with a static infrastructure
to a resource for business flexibility.
Driving business value with a
virtualized infrastructure.
Page 14
Highlights
Intellectual capital and established
processes can add the under-
standing of business challenges
necessary to create a leading-edge
deployment.
Driving business value with a
virtualized infrastructure.
Page 15
Highlights
For more information
ibm.com/services
© Copyright IBM Corporation 2007
IBM Global Services
Route 100
Somers, NY 10589
U.S.A.
Produced in the United States of America
03-07
All Rights Reserved
IBM and the IBM logo are trademarks or registered
trademarks of International Business Machines
Corporation in the United States, other countries,
or both.
Java and all Java-based trademarks are trade-
marks of Sun Microsystems, Inc. in the United
States, other countries, or both.
Linux is a registered trademark of Linus Torvalds in
the United States, other countries, or both.
Other company, product and service names may
be trademarks or service marks of others.
References in this publication to IBM products or
services do not imply that IBM intends to make them
available in all countries in which IBM operates.
GSW01064-USEN-00
May 2007
MANAGEMENT BRIEF
IMPACT OF IBM SYSTEM p SERVER VIRTUALIZATION
Transforming the IT Value Equation with POWER6 Architecture
International Technology Group
4546 El Camino Real, Suite 230 Los Altos, California 94022-1069
Telephone: (650) 949-8410 Facsimile: (650) 949-8415 Email: [email protected]
ITG
Copyright © 2007 by the International Technology Group. All rights reserved. Material, in whole or part, contained in this document may not be reproduced or distributed by any means or in any form, including original, without the prior written permission of the International Technology Group (ITG). Information has been obtained from sources assumed to be reliable and reflects conclusions at the time. This document was developed with International Business Machines Corporation (IBM) funding. Although the document may utilize publicly available material from various sources, including IBM, it does not necessarily reflect the positions of such sources on the issues addressed in this document. Material contained and conclusions presented in this document are subject to change without notice. All warranties as to the accuracy, completeness or adequacy of such material are disclaimed. There shall be no liability for errors, omissions or inadequacies in the material contained in this document or for interpretations thereof. Trademarks included in this document are the property of their respective owners.
International Technology Group i
TABLE OF CONTENTS
EXECUTIVE SUMMARY 1 What Strategy? 1 Economic Benefits 1
Overview 1 UNIX Server Costs 1 Linux Server Costs 4
Conclusions 7 STRATEGIC ROLE 8
IBM SYSTEM p SERVER 11
Overview 11 Virtualization Capabilities 11
Partitioning and Management 11 Virtual I/O Server 13
System-level Performance 14 Availability Optimization 14
Core RAS Features 14 Higher-level Capabilities 15
Autonomic Functions 15 COST PICTURE 16
UNIX Server Costs 16 Basis of Calculations 16 Cost Breakdowns 18 Cost Factors 19
Linux Server Costs 19 Basis of Calculations 19 Cost Breakdowns 21 Cost Factors 21
Costs of Downtime 22 Downtime Impacts 22 UNIX Server Comparisons 22 Linux Server Comparisons 23
SERVER COST CALCULATIONS 24 General Approach 24 Configuration Sizing 24 Cost Components 24
List of Figures 1. UNIX Server Comparisons: Five-year Operating Costs 2 2. UNIX Server Comparisons: Five-year Overall Costs
(Virtualized Scenarios include Operating and Acquisition Costs) 2 3. UNIX Server Comparisons: Five-year Operating Costs
for p570 and Equivalent Conventional Servers 3 4. UNIX Server Comparisons: Five-year Overall Costs
for p570 and Equivalent Conventional Servers (Virtualized Scenarios include Operating and Acquisition Costs) 3
5. Five-Year I/O-related Costs for Partitioned IBM System p Servers: All Virtualized Scenarios 4
6. Linux Server Comparisons: Five-year Operating Costs 5
International Technology Group ii
7. Linux Server Comparisons: Five-year Overall Costs (Virtualized Scenarios include Acquisition Costs) 5
8. UNIX Server Comparisons: Five-year Costs of Downtime 6 9. Linux Server Comparisons: Five-Year Costs of Downtime 6 10. IT Expenditure by Fortune 1000 Companies in 2006: Conventional Breakdown 8 11. IT Expenditure by Fortune 1000 Companies in 2006: Alternative Breakdown 9 12. Applications Expenditure by Fortune 1000 Companies in 2006 9 13. Key IBM System p Virtualization Capabilities 12 14. Dedicated Adapter and Virtual I/O Server Configurations: Examples 13 15. IBM System p Autonomic Functions 15 16. UNIX Server Profiles Summary 17 17. Five-Year UNIX Server Costs Detail 18 18. UNIX Server Comparisons: Factors Resulting in Lower Operating Costs
for Virtualized Scenarios (All Servers) 19 19. UNIX Server Comparisons: Factors Resulting in Lower Operating Costs
for p570 Servers and Conventional Equivalents 19 20. Linux Server Installation Profiles Summary 20 21. Five-Year Linux Server Costs Detail 21 22. Linux Server Comparisons: Factors Resulting in Lower Operating Costs
for Virtualized Scenarios 21
International Technology Group 1
EXECUTIVE SUMMARY
What Strategy?
Virtualization is no longer an “emerging” technology. Its potential to increase the efficiency of IT infrastructures has become widely recognized. Organization-wide improvements in overall capacity utilization and reductions in the costs of acquiring, managing and supporting server and storage platforms have become realistic opportunities.
Yet, in most organizations, use of virtualization is still at any early stage. The focus is on small-scale projects. One reason for this is that, while many organizations recognize the potential of virtualization in general terms, they are unclear as to what the scope and goals of virtualization strategy should be.
Key questions are raised. Should virtualization be implemented only for new projects, or applied to existing infrastructures? Should opportunities be pursued on a case-by-case basis, or targeted as a broader goal of IT strategy? What level of investment should be made, for what return? What priority should be given to virtualization initiatives relative to the many other claims on IT resources?
Before these questions can be answered, a broader understanding of the costs and benefits of virtualization at the enterprise level is needed. This report marks a first step in this process.
Economic Benefits
Overview
This report focuses on the potential economic benefits of large-scale virtualization for two subsets of the large organization IT environment – UNIX and Linux servers.
Specifically, it deals with the potential benefits of strategies that effectively exploit the virtualization strengths of the IBM System p platform. It addresses established System p strengths, as well as the new capabilities announced by IBM with the introduction of its POWER6 processor-based p570 servers in May 2007.
Potential benefits are illustrated by three composite profiles of large financial services, manufacturing and retail companies. For each profile, two sets of scenarios are presented: (1) conventional scenarios built around diverse, multivendor bases of UNIX and Intel-based Linux servers; and (2) virtualized scenarios in which the same applications run on System p servers leveraging the full potential of virtualization.
For each profile and scenario, costs are then compared. Results are summarized below.
UNIX Server Costs
Five-year server operating costs for System p virtualized scenarios including POWER6 processor-based p570 and POWER5+ processor-based System p5 servers averaged 72 percent less than those for conventional scenarios.
Operating costs, in this context, include hardware maintenance; update subscriptions and support for systems and database software; personnel for system administration-related functions; and facilities costs for data center occupancy, power and cooling.
Even allowing for the costs of acquiring new System p hardware and systems software required to realize these savings, overall costs were still significantly lower. If acquisition costs are included in virtualized scenarios, five-year costs for these averaged 66 percent less than those for conventional scenarios.
International Technology Group 2
Figures 1 and 2 summarize results.
Figure 1 UNIX Server Comparisons: Five-year Operating Costs
Figure 2 UNIX Server Comparisons: Five-year Overall Costs
(Virtualized Scenarios include Operating and Acquisition Costs)
Detailed breakdowns of costs, along with explanations of variations between scenarios for these comparisons are provided in figures 17 and 18 respectively in the Cost Picture section of this report.
Cost reductions were enabled through multiple System p virtualization capabilities. Dynamic logical partitions (LPARs), along with Workload Partitions (WPARs) combine with effective system and workload management facilities to enable high levels of server consolidation and capacity utilization.
0 10 20 30 40 50 60
Virtualized
Conventional
Virtualized
Conventional
Virtualized
Conventional
$ Millions
MaintenanceSystem software supportDatabase software supportPersonnelFacilities
FINANCIAL SERVICES COMPANY SCENARIOS
56.3
14.3
25.1
7.0
20.9
7.0
MANUFACTURING COMPANY SCENARIOS
RETAIL COMPANY SCENARIOS
0 10 20 30 40 50 60
Virtualized
Conventional
Virtualized
Conventional
Virtualized
Conventional
$ Millions
HardwareSystem softwareMaintenanceSystem software supportDatabase software supportPersonnelFacilities
FINANCIAL SERVICES COMPANY SCENARIOS
56.3
17.8
25.1
8.6
20.9
8.2
MANUFACTURING COMPANY SCENARIOS
RETAIL COMPANY SCENARIOS
International Technology Group 3
The comparisons in figures 1 and 2 are between conventional scenarios and virtualized scenarios, which include p570 as well as System p5 servers.
The new POWER6 processor-based System p 570 servers, however, offer significant improvements in performance and virtualization capability. If costs of p570 models in virtualized scenarios are compared with those of equivalent servers in conventional scenarios, disparities widen. Five-year operating costs for p570 servers averaged 79 percent less than those for conventional equivalents.
If acquisition costs for p570 servers are included in comparisons, five-year costs averaged 72 percent less than those for conventional equivalents. Figures 3 and 4 summarize results.
Figure 3 UNIX Server Comparisons: Five-year Operating Costs
for p570 and Equivalent Conventional Servers
Figure 4 UNIX Server Comparisons: Five-year Overall Costs for p570 and Equivalent Co nventional
Servers (Virtualized Scenarios include Operating and Acquisition Costs)
0 10 20 30 40 50
Virtualized
Conventional
Virtualized
Conventional
Virtualized
Conventional
$ Millions
MaintenanceSystem software supportDatabase software supportPersonnelFacilities
FINANCIAL SERVICES COMPANY SCENARIOS
41.7
9.0
20.5
4.4
16.7
3.4
MANUFACTURING COMPANY SCENARIOS
RETAIL COMPANY SCENARIOS
0 10 20 30 40 50
Virtualized
Conventional
Virtualized
Conventional
Virtualized
Conventional
$ Millions
HardwareSystem softwareMaintenanceSystem software supportDatabase software supportPersonnelFacilities
FINANCIAL SERVICES COMPANY SCENARIOS
41.7
12.0
20.5
5.9
16.7
4.4
MANUFACTURING COMPANY SCENARIOS
RETAIL COMPANY SCENARIOS
International Technology Group 4
Detailed breakdowns of costs, along with explanations of variations between scenarios for these comparisons are provided in figures 17 and 19 respectively in the Cost Picture section of this report.
Lower costs for p570 servers are due to virtualization capabilities supported on System p5 servers, as well as to higher levels of performance and new POWER6 processor-specific capabilities such as Shared Dedicated Capacity, which enable further improvements in system-level capacity utilization.
For p570 as well as System p5 servers, significant economies were realized in I/O-related costs, including local area network (LAN) and storage area network (SAN) adapters, as well physical infrastructures supporting these through use of System p Virtual I/O Server technology. Physical infrastructures include transceivers, structured cabling and switches.
In virtualized scenarios, use of Virtual I/O Servers resulted in I/O-related costs for servers employing LPARs or both that averaged 65 percent less than would have been the case if dedicated adapters were employed. Figure 5 summarizes results.
Figure 5 Five-Year I/O-related Costs for Partitioned IBM System p Servers:
All Virtualized Scenarios
For both sets of comparisons, full time equivalent (FTE) staffing levels for system administration-related functions such as asset, capacity, change, configuration and performance management are significantly lower for virtualized scenarios. Personnel costs for these functions are correspondingly less.
This is due to reductions in numbers of physical servers; reduced diversity of hardware and software platforms; replacement of older-technology servers and systems software with latest-generation System p hardware and software platforms; improved management tools and practices, including increased automation; and use of System p virtualization capabilities.
Linux Server Costs
In addition to the AIX operating system, System p servers may also run the major Linux distributions in native mode. The new IBM System p Application Virtual Environment (System p AVE) allows x86 Linux applications to run on the System p platform without modification. Many of the same virtualization capabilities are supported as for AIX, making this platform a candidate for Linux server consolidation.
To address this potential, comparisons were made for different sets of Linux-based applications for all three composite profiles. Comparisons are based on mixes of business applications as well as Web and intranet serving, file serving, software development and other functions.
Two sets of scenarios were compared. In conventional scenarios, applications are deployed on one- to four-way Intel-based servers. In virtualized scenarios, System p 570 and System p5 servers are employed.
Figures 6 and 7 summarize results.
0 100 200 300 400 500 600 700 800 900 1,000
Virtual I/O ServerConfigurations
Dedicated AdaptersConfigurations
LAN Adapters SAN Adapters Infrastructure
912
324
$ Thousands
International Technology Group 5
Figure 6 Linux Server Comparisons: Five-year Operating Costs
Figure 7
Linux Server Comparisons: Five-year Overall Costs (Virtualized Scenarios include Acquisition Costs )
Five-year operating costs for System p Linux server scenarios averaged 67 percent less than those for conventional scenarios. If acquisition costs are included in System p scenarios, five-year costs for these averaged 57 percent less than those for conventional scenarios.
In this presentation, operating costs again include hardware maintenance; update subscriptions and support for systems and database software; personnel for system administration-related functions; and facilities costs for data center occupancy, power and cooling. Database costs are for Oracle 10g.
Detailed breakdowns of costs, along with explanations of variations between scenarios for these comparisons are provided in figures 21 and 22 in the Cost Picture section of this report.
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Virtualized
Intel-based Conventional
Virtualized
Intel-based Conventional
Virtualized
Intel-based Conventional
$ Millions
MaintenanceDatabase software supportPersonnelFacilities
FINANCIAL SERVICES COMPANY SCENARIOS
3.87
1.17
0.97
0.31
1.69
0.66
MANUFACTURING COMPANY SCENARIOS
RETAIL COMPANY SCENARIOS
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0
Virtualized
Intel-based Conventional
Virtualized
Intel-based Conventional
Virtualized
Intel-based Conventional
$ Millions
HardwareMaintenanceDatabase software supportPersonnelFacilities
FINANCIAL SERVICES COMPANY SCENARIOS
3.87
1.55
0.97
0.41
1.69
0.84
MANUFACTURING COMPANY SCENARIOS
RETAIL COMPANY SCENARIOS
International Technology Group 6
Factors causing lower costs for virtualized scenarios are generally similar to those for UNIX server comparisons. The same caveats as to definitions of system administration-related functions apply.
Costs of Downtime
Because of higher levels of availability are enabled by virtualized environments, there are also significant differences in costs of downtime – meaning costs incurred by organizations due to disruptions affecting critical business processes – between conventional and virtualized scenarios.
For UNIX serving, five-year costs of downtime for System p virtualized scenarios ranged from 41 percent to 59 percent less, and averaged 48 percent less than those for conventional scenarios.
For Linux serving, System p virtualized scenarios ranged from 83 percent to 89 percent less, and averaged 85 percent less than those for Intel-based conventional scenarios. Figures 8 and 9 summarize results.
Figure 8 UNIX Server Comparisons: Five-year Costs of Downtime
COMPANY FINANCIAL SERVICES MANUFACTURING RETAIL
Applications Online banking & brokerage, CRM financial services, various
ERP, SCM, CRM, BI, CFM, SEM, PLM, procurement
SCM, logistics, procurement, eCommerce, CRM
CONVENTIONAL SCENARIOS
Availability levels
Five-year costs ($000)
99.3% – 99.92%
25,649
99.75% – 99.9%
36,950
98.25% – 99.9%
13,419
VIRTUALIZED SCENARIOS
Availability levels
Five-year costs ($000)
99.7% – 99.95%
10,505
99.9% – 99.96%
21,444
98.5% – 99.95%
7,882
Figure 9 Linux Server Comparisons: Five-Year Costs of Downtime
COMPANY FINANCIAL SERVICES MANUFACTURING RETAIL
CONVENTIONAL SCENARIOS
Availability levels
Five-year costs ($000)
99.32% – 99.83%
1,858
98.05% – 99.61%
3,299
99.14% – 99.77%
2,588
VIRTUALIZED SCENARIOS
Availability levels
Five-year costs ($000)
99.90% – 99.98%
213
99.85% – 99.95%
557
99.73% – 99.98%
420
In these figures, availability percentages reflect hours of system-level outages relative to annual hours of operation of companies, or business areas within companies that are supported by specific applications. The basis of these calculations is detailed in the Cost Picture section of this report.
Costs of downtime for UNIX server scenarios represent lost operating profit (for the financial services profile) or gross profit (for the manufacturing and retail companies). For Linux server scenarios, costs of downtime represent lost productivity. Both sets of calculations are based on industry- and organization-specific values for the effects of outages. The basis of these calculations is detailed in the Cost Picture section of this report.
Lower costs of downtime for virtualized scenarios are due to multiple factors that reduce the frequency and duration of planned as well as unplanned outages. These include reliability, availability and serviceability (RAS) features of System p hardware and the AIX operating system, and the IBM High Availability Cluster Multi Processing (HACMP) failover solution.
International Technology Group 7
LPARs also assist in avoiding planned outages by allowing users to upgrade or modify software without taking systems offline. Two new capabilities – Live Partition Mobility and Live Application Mobility – make it possible to transfer partitions between servers if these need to be shut down for hardware-related reasons.
Live Partition Mobility is supported for AIX and Linux on p570 servers, and enables transfers to occur with no application downtime. It is designed to support applications requiring continuous availability. Live Application Mobility involves a brief interruption of service. It is supported on POWER6 processor-based as well as earlier POWER4 and POWER5 processor-based models running AIX 6.
Effective system and workload management capabilities further reduce risks of bottlenecks and outages caused by workload spikes and operational errors.
Conclusions
Realization of the gains described in this report would require investments in a number of areas, such as migration of applications and workloads, organizational change and staff retraining, which are not addressed by this report. Actual user savings may vary widely in practice.
The principle is nevertheless demonstrated: efficient virtualized server infrastructures built around the System p offer the potential for major improvements in IT cost structures and availability levels. Even if organizations do not realize large-scale, short-term turnover of server bases, the phased creation of such infrastructures over time represents a new, highly attractive opportunity to increase IT cost-effectiveness.
There is a particularly significant opportunity to support large-scale Linux deployment. Use of the System p platform could enable organizations to put in place efficient, scalable and robust server infrastructures at an early stage of deployment. It would become possible to avoid many of the challenges of infrastructure fragmentation, problematic availability and management overhead that, all too often, have been characteristic of x86 server bases.
Linux and virtualization have both been described as “disruptive” or “transformative” technologies. Less well recognized, however, is that the combination of these may represent a force for change that is broader and more powerful than either individually.
Most industry observers believe that the potential of Linux virtualization will not be realized for years. But that is the case for the x86 world. The System p server offers that potential now. Which means, for many organizations, a new field of opportunities may be realized sooner rather than later.
International Technology Group 8
STRATEGIC ROLE Beyond the quantifiable cost and benefit issues that are the focus of this report, a larger question is raised: what role might virtualization play in meeting the central challenges that IT strategy must address in the 21st century?
The large organization IT community is in a conservative mood. After aggressive deployment of new applications such as enterprise resource planning (ERP), customer relationship management (CRM) and eBusiness systems in the 1990s, the focus has shifted to solutions that offer more incremental forms of business advantage. Integration of existing environments, rather than delivery of new functionality, dominates many corporate IT agendas.
Structural factors have contributed to this situation. Past deployments have created layers of complexity that undermine overall IT effectiveness. Resources have increasingly been diverted into managing and maintaining underlying platforms, rather than into the higher value-added processes of application delivery and business transformation.
The extent to which this has occurred has been obscured, all too often, by the way in which organizations account for IT expenditure. Conventional categories such as hardware, systems software, middleware, applications, networks, external services, personnel and facilities may be useful for budgetary purposes. They do not necessarily provide insight into underlying cost structures.
The implications may be simply illustrated. Figure 10 shows a conventional breakdown of IT expenditures by Fortune 1000 companies during 2006.
Figure 10 IT Expenditure by Fortune 1000 Companies in 2006: Conventional Breakdown
Alternatively, IT expenditure may be divided into two main categories: applications and infrastructures. If Fortune 1000 outlays during 2006 are broken down in this manner, the results are as shown in figure 11.
TOTAL: $401.58 BILLION
Servers/storage15.0%
PCs/clients6.1%
Networks6.8%
External services10.1%
Facilities4.5%
Other2.3%
Personnel36.7%
Applications software12.3%
Systems software6.2%
International Technology Group 9
Figure 11 IT Expenditure by Fortune 1000 Companies in 2006: Alternative Breakdown
In this presentation, personnel, services and facilities costs, as well as those for hardware and software, are assigned to functional categories. For example, “applications” costs include applications software license and support fees, along with costs of in-house personnel and external services for application development, deployment and maintenance.
A striking conclusion emerges: less than a quarter of overall Fortune 1000 IT expenditures in 2006 were for applications. More than three-quarters went into infrastructures.
Moreover, the majority of applications expenditure is accounted for by update and support fees for installed applications, changes and enhancements to these, and tools and programming for systems integration and interoperability. As figure 12 shows, an average of only approximately seven percent of overall IT expenditure went into the delivery of new application functionality.
Figure 12 Applications Expenditure by Fortune 1000 Companies in 2006
TOTAL: $401.58 BILLION
APPLICATIONS23.4%
Servers/storage33.9%
Middleware15.1%
Networks13.7%
End-user computing11.4%
Other2.5%
INFRASTRUCTURES76.6%
TOTAL: $401.58 BILLION
Other2.5%
Servers/storage33.9%
Applications23.4%
End-user computing11.4%
Integration/interoperability7.4%
Middleware15.1%
Networks13.7%
Changes/enhancements4.2%
Update & support fees4.5%
New application functionality7.3%
International Technology Group 10
A key principle should be highlighted. Applications are the direct source of business value delivered by IT. Users interact with, and business processes are enabled by applications. Underlying IT resources such as databases, middleware, servers, storage, PCs and networks are simply the delivery mechanisms for these. Yet, the majority of resources are expended on delivery mechanisms.
The impact of infrastructure fragmentation, however, extends beyond simple diversion of resources. Complexity of underlying infrastructures has materially increased the cost and risk involved in major new application deployments, and has encouraged a focus on maintenance and incremental enhancement of existing applications.
This situation has inevitably discouraged large-scale application initiatives. It has also helped to create a perception among many corporate executives that IT investment is a comparatively high-risk, low-reward proposition, and that organizational resources should be focused elsewhere.
It is from this perspective that the strategic significance of virtualization should be viewed. Even small reductions in overall infrastructure complexity could enable organizations to significantly increase the resources devoted to new application initiatives, and to reduce costs, risks and delivery times for these.
Virtualization initiatives may thus act as a catalyst not only for major shifts in IT cost structures, but also for broader, transformational increases in IT effectiveness.
Organizations are often wary of “quantum leaps” in technology. But a quantum leap in the organization and operation of server infrastructures has nevertheless become possible.
International Technology Group 11
IBM SYSTEM p SERVER
Overview
The virtualization capabilities of the IBM System p server form part of a larger system design that is also characterized by high levels of system-level performance, advanced virtualization capabilities, optimization for availability as well as other variables of service quality, and extensive autonomic capability – meaning the application of artificial intelligence technologies to IT administration tasks.
The potential of the System p server for IT cost reduction, which is illustrated in this report, is a function of all of these capabilities. The central strength of the platform is that these capabilities are designed into the core system architecture, and integrated and optimized in a mutually reinforcing manner. The overall impact in terms of operating efficiency is such that “the whole is more than the sum of the parts.”
Virtualization Capabilities
Partitioning and Management
System p capabilities in this area include the following:
• Partitioning. The System p platform supports three complementary forms of partitioning. Hardware-based LPARs are implemented through a firmware-based hypervisor, and enable processor, memory and Ethernet resources to be shared between instances of AIX and Linux. Up to 64 LPARs with one or more cores each are supported.
Advanced POWER Virtualization (APV), an optional System p feature, allows I/O resources to be shared and dynamically re-allocated between instances of AIX 5.3 and later, and Linux. Partitions may be configured in increments as small as one tenth of a core. This capability is commonly employed to improve load balancing for large, complex workloads.
WPARs, which are supported for AIX 6, provide a further level of capability. WPARs allow users to create multiple software-based partitions on top of a single AIX instance. This approach enables high levels of flexibility and capacity utilization for applications executing heterogeneous workloads, and simplifies patching as well as other operating system maintenance tasks.
These technologies, separately and – to an even greater extent – combined, allow organizations to consolidate multiple applications, as well as multiple components of the same application (e.g., database, application and Web serving) that would otherwise require dedicated servers.
• System and workload management. Partitioning techniques enable consolidation, creating the potential for significantly higher levels of overall capacity utilization than may be realized with dedicated servers.
The extent to which this potential will be realized in practice, however, depends heavily on the mechanisms that allocate system resources between, and monitor and control workload execution processes across partitions. If these are ineffective, a high proportion of system capacity may be idle at any given time. Surges in workloads running in individual partitions may also create bottlenecks if additional capacity is not available in a timely manner.
One of the core strengths of the System p platform is that highly effective management facilities are implemented at multiple levels, and closely integrated and optimized within the overall system environment. This is illustrated in figure 13.
International Technology Group 12
Figure 13 Key IBM System p Virtualization Capabilities
For p570 servers, an additional capability has been announced by IBM called Shared Dedicated Capacity, which provides options to donate excess cycles from dedicated processors to a shared pool
Higher-level management capabilities are provided by the IBM Systems Director family of tools and by IBM Tivoli enterprise management solutions.
System p management facilities also address a set of requirements that have proved important in encouraging large-scale adoption of virtualization: the ability to track usage of shared system resources in a manner that enables organizations to implement effective chargeback procedures.
AIX contains an extensive suite of system-level accounting features that enable administrators to collect statistics on usage of resources such as processors, memory, disks and adapters by application, process, transaction or other variables, or combinations of these.
In many organizations, one of the reasons for continuing to employ dedicated servers is that this approach enables the business units that control the applications that run on these servers to track costs in a simple and non-controversial manner. If, however, costs of shared server platforms can be transparently and credibly determined, and allocated accordingly, a significant organizational obstacle to large-scale server consolidation may be removed.
AIX
SYSTEM MANAGEMENT SERVICES
IBM Systems Director
Shared Processor Pool Shared Dedicated Capacity
HYPERVISOR
Integrated Virtual Ethernet
Physical Adapters
VIRTUAL I/O SERVER
Virtual Ethernet Adapters Virtual SCSI Adapters
LPAR
LPAR
LPAR
WPARs
LPAR
WPARs
Virtual LAN
International Technology Group 13
Virtual I/O Server
A second key set of System p virtualization capabilities is implemented in the Virtual I/O Server, an LPAR-based appliance that allows for the creation of virtual Ethernet and SCSI adapters.
The principal benefit of the Virtual I/O Server is that it allows operating system instances running in multiple partitions to share a common pool of LAN adapters, as well as Fiber Channel, SCSI and RAID devices. It is not necessary to dedicate adapters to individual partitions. As figure 14 illustrates, the number of physical adapters may be significantly reduced.
Figure 14 Dedicated Adapter and Virtual I/O Server Configurations: Examples
This approach may not be appropriate for all partitions. For most workloads, however, it offers the potential for significant savings in adapter and related LAN and SAN infrastructure costs.
Virtual I/O Servers interface to a virtual LAN or multiple virtual LANs that provide high-speed interconnection between LPARs within the System p environment. Virtual LAN capability further reduces network complexity and vulnerability, and may significantly reduce throughput times for interaction between LPAR-based systems, as well as for replication and other data movement processes.
A more basic form of I/O virtualization is provided for p570 servers through Integrated Virtual Ethernet. This hypervisor-based facility enables sharing of Ethernet adapters without use of a Virtual I/O Server. It is designed for comparatively light workloads that do not require high levels of I/O throughput, such as those generated by development and test systems.
Adapters
DEDICATED ADAPTER CONFIGURATION
Partitions
1
2
3
6
5
4
Partitions
VIRTUAL I/O SERVER CONFIGURATION
1
2
3
6
5
4
Adapters
VIRTUAL I/O SERVER
International Technology Group 14
System-level Performance
The System p platform benefits from high levels of performance delivered by the IBM POWER RISC design. Current POWER5+ processors include dual- and quad-core units with clock speeds of 1.5 GHz to 2.3 GHz, while p570 servers employ dual-core 3.5 GHz, 4.2 GHz and 4.7 GHz processors.
Processor performance, however, is only part of the picture. System-level performance potential has been optimized at all levels of design and implementation – including microelectronics, module- and subsystem-level components, internal communications, I/O, and system-level hardware and software.
Key System p capabilities include highly effective compiler- and operating system-level performance acceleration, including chip symmetric multithreading; low levels of symmetric multiprocessing (SMP) overhead; and extensive system-level integration and optimization of all performance-related features.
Availability Optimization
Core RAS Features
The System p server benefits from a wide range of hardware- and software-based reliability, availability and serviceability features. These are designed to reduce the potential for unplanned outages, as well as to limit the frequency and duration of planned outages for tasks such as configuration upgrades and scheduled maintenance.
Core RAS features include the following:
• Basic capabilities include high levels of component reliability and redundancy, along with pervasive monitoring, diagnostic, and fault isolation and resolution facilities. These are built into processors, main memory, cache and packaging modules, as well as into all major hardware components. In many cases, multiple layers of protection and self-test are implemented.
Key functionality is delivered by IBM-developed Chipkill and First Failure Data Capture (FFDC) technologies. Chipkill, which performs error checking for memory devices, is regarded as more reliable than conventional error correction code (ECC) techniques. FFDC employs thousands of embedded sensors that identify and report failures to a separately powered Service Processor, which also monitors environmental conditions.
FFDC architecture forms the basis for predictive failure analysis functions that identify potential as well as actual failures throughout the system. In both cases, the Service Processor can automatically notify system administrators or contact an IBM Support Center directly (“call home” service) to report events requiring service intervention.
• Failure masking capabilities prevent outages in case failures do occur. For example, processors may be automatically disabled if they begin to malfunction, and standby processors may be activated without interrupting operations. Dynamic LPARs facilitate this process.
Concurrent maintenance (“hot plugging”) functions and dynamic also reduce requirements for planned outages. LPARs reduce them further. Systems software and application upgrades, for example, may be performed in one LPAR while the original system continues operating in another. Software may be copied to and modified in LPARs. Backups may be executed concurrently with online processing.
System p RAS features are implemented at multiple levels, including LPARs (availability-related functions are built into the System p hypervisor), the Virtual I/O Server (for adapters) and system-level management facilities. These capabilities draw extensively on mainframe high availability design concepts and technologies.
International Technology Group 15
Higher-level Capabilities
Higher-level capabilities include the following:
• Live Partition Mobility and Live Application Mobility are new capabilities that enable users to move partitions between systems with no application downtime, or limited downtime respectively.
Live Partition Mobility is designed primarily for organizations that need to shut down a System p server for maintenance, upgrades and other functions, but cannot afford to take business-critical production systems offline. These may be simply transferred to another server, and returned after the original server is restarted.
The only interruption of service would be due to network latency. If sufficient bandwidth was available, a delay of – at most – a few seconds could typically be expected. Live Application Mobility involves longer delays (e.g., 20 seconds), and would normally be employed for less availability-sensitive applications.
• High Availability Cluster Multiprocessing is IBM’s principal solution for System p failover clustering. Organizations employing HACMP have achieved mainframe-class availability levels even for highly demanding, complex workloads.
If HACMP/Extended Distance (HACMP/XD) is employed, failover may occur at distances of up to 300 kilometers if sufficient wide area network bandwidth is available.
These solutions provide a spectrum of capability designed to meet a wide range of user needs. Live Partition Mobility and Live Application Mobility, for example, do not require use of HACMP or other cluster solutions that may reduce costs while avoiding disruptions of service. It can be expected that users will continue to employ HACMP to protect against unplanned outages, and for disaster recovery.
Autonomic Functions
The System p platform benefits from one of the most advanced implementations of autonomic capability within the IBM product line. Autonomic functions, which are grouped by IBM into four categories – self-configuring, self-optimizing, self-protecting and self-healing – are summarized in figure 15.
Figure 15 IBM System p Autonomic Functions
SELF-CONFIGURING SELF-OPTIMIZING SELF-PROTECTING SELF-HEALING
Virtual IP address IP multipath routing Microcode discovery Hot-swappable disks Hot-swap PCI Wireless/pervasive configuration TCP explicit congestion notification
Static LPAR Dynamic LPAR Workload manager enhancement Extended memory allocator RSCT technology PSSP cluster management
Kerberos V5 Authentication Self-protecting kernel SecureWay LDAP directory integration SSL management Digital Certificates Encryption
Multiple default gateways Automatic system hang recovery Automatic dump analysis EtherChannel automatic failover Processor failure detection & failover First failure data capture Chipkill ECC Memory Dynamic bit steering Memory scrubbing Automatic dynamic deallocation Electronic Service Agent
In addition to augmenting the effectiveness of System p RAS capabilities, autonomic functions contribute to higher levels of system administration productivity by streamlining and automating tasks that would otherwise require extensive manual intervention.
International Technology Group 16
COST PICTURE
UNIX Server Costs
Basis of Calculations
The UNIX server cost comparisons presented in this report are based on three composite profiles of large company installations employing a variety of Hewlett-Packard, IBM, Sun Microsystems and other servers running variants of the UNIX operating system other than Linux.
Profiles were constructed using data on application portfolios, server bases, configurations, utilization and service levels, staffing and other variables supplied by 16 companies in the same industries and approximate size ranges, with generally similar business profiles.
Two sets of scenarios were then developed:
1. Conventional scenarios are based on data reported by the 16 companies, and are built around diverse multivendor server bases that include different technology generations and systems software versions. Conventional management and operating practices are employed.
2. Virtualized scenarios are for the same applications and workloads as conventional scenarios. Configurations for these scenarios were developed on a case-by-case basis within major system groups. Where appropriate, multiple applications, application instances, or both were configured on the same physical servers using LPARs, WPARs, or combinations of these.
All profile scenarios include database, application and, where appropriate, Web and intranet servers; and production systems as well as non-production instances for such functions as development, test, quality assurance and training. Clustered failover configurations are employed in conventional and virtualized scenarios for systems requiring high levels of availability.
Financial services company scenarios include a variety of custom and packaged software solutions. Manufacturing company scenarios are built primarily around SAP AG and complementary third-party applications. Retail company scenarios include a mix of applications from EXE Technologies, i2 Technologies, JDA Software, Oracle (including PeopleSoft and Retek) and other vendors.
Profile installations are summarized in figure 16.
International Technology Group 17
Figure 16 UNIX Server Profiles Summary
COMPANY FINANCIAL SERVICES MANUFACTURING RETAIL
Business Profile $325 billion assets $16 billion revenues 12 million customers 1,500 branches 50,000 employees
$27 billion revenues Consumer products 30 manufacturing plants 25 distribution centers 50,000 employees
$30 billion revenues Specialty chain 900+ outlets 10 distribution centers 125,000 employees
Focus of Comparisons 64 major applications System groups: • Corporate systems • Retail banking • Commercial banking • Financial services • Online systems • Intranet infrastructure
27 major applications System groups: • Core ERP systems • Supply chain management • Product management • Business intelligence • eProcurement • CRM
35 major applications System groups: • Corporate systems • Supply chain management • Logistics & transportation • Marketing, sales & service • Business intelligence • eCommerce
CONVENTIONAL SCENARIOS
Servers Hewlett-Packard Superdome, rx8640, rx8620, rx7620, rx6600, rx5670, rx4640, rx2620, rx2600, rp8420, rp8400, N4000, various Hewlett-Packard (Alpha) GS80, ES45, ES40 IBM p690, p680, p670, p660, p650, p640, p630, p615, p610, p5-570, p5-550, p5-510, p5-505, p5-185 Sun Microsystems E25K, E10K, E6800, E4900, E4800, E4500, E2900, E450, V890, V880, V490, V480, V440, V250, V240, V40Z, V20Z, X4600, X4200, X4100, various Silicon Graphics Altix 450
Total: 285 servers
Hewlett-Packard Superdome, rx7640, rx7620, rx6600, rx4640, rx2620, rp8420, rp5470, rp440 IBM p690, p670, p660, p650, p630, p615, p5-570, p5-550, p5-520, p5-510 Sun Microsystems E20K, E15K, E4800, V890, V880, V490, V480, V250, V40Z, V20Z, X4200, X4100, various
Total: 85 servers
Hewlett-Packard rx8640, rx8620, rx7620, rx5670, rx4640, rx2620, rx2600, rx1600, rp8400 IBM p690, p680, p670, p650, p630, p615, p5-570, p5-550, p5-510, RS/6000 Sun Microsystems E15K, E6800, E4900, E4800, E4500, V890, V490, V480, V440, V250, V240, V40Z, V20Z, X4600, X4200, various
Total: 112 servers
Personnel 34 FTEs 14 FTEs 16 FTEs
VIRTUALIZED SCENARIOS
Servers 13 x POWER6 p570 4 x p5-550Q, 3 x p5-520Q 2 x p5-510Q, 24 x p5-505Q 22 x p5-505, 12 x p5-185 Totals: 80 servers 42 LPARs 82 WPARs
5 x POWER6 p570 5 x p5-550Q, 2 x p5-520Q 13 x p5-505Q, 5 x p5-505 Totals: 30 servers 28 LPARs 41 WPARs
7 x POWER6 p570 2 x p5-550Q, 3 x p5-520Q 14 x p5-505Q, 15 x p5-505 8 x p5-185 Totals: 49 servers 34 LPARs 56 WPARs
Personnel 17 FTEs 9 FTEs 10 FTEs
International Technology Group 18
Cost Breakdowns
For conventional scenarios, calculations are for operating costs only. These include hardware maintenance, update and support subscriptions for systems and database software, along with personnel and facilities costs. For virtualized scenarios, calculations are for operating costs, which include the same components, along with costs of hardware and systems software acquisition for new System p servers.
Detailed breakdowns of calculations are shown in figure 17. Costs for p570 servers in virtualized scenarios and equivalent servers in conventional scenarios are shown separately.
Figure 17 Five-Year UNIX Server Costs Detail
COMPANY FINANCIAL SERVICES MANUFACTURING RETAIL
CONVENTIONAL SCENARIOS – ALL SERVERS ($000)
Maintenance 10,722.9 6,401.4 2,510.4
Systems software support 2,370.6 1,366.2 1,371.6
Database software support 25,264.2 9,922.3 8,714.3
Personnel 16,529.0 6,806.1 7,778.4
Facilities 1,393.9 584.3 498.1
TOTAL OPERATING COSTS 56,280.6 25,080.3 20,872.8
VIRTUALIZED SCENARIOS – ALL SERVERS ($000)
Hardware 3,137.3 1,445.2 1,123.0
Systems Software 340.2 148.9 98.4
Acquisition costs (Subtotal) 3,477.5 1,594.1 1,221.4
Maintenance 931.2 401.7 362.6
Systems software support 1,094.3 585.0 348.6
Database software support 3,590.6 1,480.1 1,221.5
Personnel 8,264.5 4,375.3 4,861.5
Facilities 419.7 206.1 167.7
Operating costs (Subtotal) 14,300.3 7,048.2 6,961.9
TOTAL 17,777.8 8,642.3 8,183.3
CONVENTIONAL SCENARIOS – p570 EQUIVALENTS ($000)
Maintenance 7,960.8 5,095.6 2,142.7
Systems software support 1,736.6 1,105.4 1,019.6
Database software support 21,334.3 9,922.3 8,293.3
Personnel 9,723.0 3,889.2 4,861.5
Facilities 920.4 465.2 340.6
TOTAL OPERATING COSTS 41,675.1 20,477.7 16,657.7
VIRTUALIZED SCENARIOS – p570 SERVERS ($000)
Hardware 2,693.5 1,319.1 880.6
Systems software 248.7 140.2 76.4
Acquisition costs (Subtotal) 2,942.2 1,459.3 957.0
Maintenance 785.8 356.7 268.1
Systems software support 774.5 529.0 239.9
Database software support 2,861.4 1,480.1 888.0
Personnel 4,375.3 1,944.6 1,944.6
Facilities 216.1 83.6 84.2
Operating costs (Subtotal) 9,013.1 4,394.0 3,424.8
TOTAL 11,955.3 5,853.3 4,381.8
International Technology Group 19
For all scenarios, database costs are for Oracle 10g. Calculations do not include initial license costs. It is assumed that these are covered by existing customer licensing arrangements.
All costs, other than for personnel and facilities, are based on discounted vendor “street” prices for the products and services included in scenarios. Personnel costs are for the numbers of FTE UNIX system administrators shown in figure 16.
Facilities costs are for data center occupancy, power and cooling equipment, electricity and related operating costs, and are based on U.S. industry averages and norms. Additional detail on the basis of these calculations may be found in the Server Cost Calculations section this report.
Cost Factors
Lower five-year operating costs for virtualized scenarios relative to conventional scenarios in profile-based comparisons are due to the factors summarized in figures 18 (all servers) and 19 (p570 servers and conventional equivalents).
Figure 18 UNIX Server Comparisons: Factors Resulting in Lower Operating Costs
for Virtualized Scenarios (All Servers)
Categories Savings Factors
Maintenance 86% - 94% Fewer, newer servers reduce maintenance contract costs.
Software support 82% - 84% Fewer software copies & CPUs result in lower license, update & support costs.
Personnel 36% - 50% Fewer physical servers, reduced diversity & improved automation reduce system administration-related personnel costs.
Facilities 65% - 70% Fewer physical servers, smaller footprints, & greater energy efficiency reduce data center occupancy, power & cooling costs.
Average 72%
Figure 19 UNIX Server Comparisons: Factors Resulting in Lower Operating Costs
for p570 Servers and Conventional Equivalents
Categories Savings Factors
Maintenance 87% - 93% Fewer, newer servers reduce maintenance contract costs.
Software support 82% - 88% Fewer software copies & CPUs result in lower license, update & support costs.
Personnel 50% - 60% Fewer physical servers, reduced diversity & improved automation reduce system administration-related personnel costs.
Facilities 75% - 82% Fewer physical servers, smaller footprints, & greater energy efficiency reduce data center occupancy, power & cooling costs.
Average 79%
Linux Server Costs
Basis of Calculations
The Linux server cost comparisons presented in this report are based on the same company business profiles employed for UNIX server comparisons. Installation profiles, however, are different. Application portfolios, server bases and staffing levels were developed using data from 21 companies, including in some cases companies that did not supply UNIX server data.
International Technology Group 20
Linux installation profiles are summarized in figure 20.
Figure 20 Linux Server Installation Profiles Summary
COMPANY FINANCIAL SERVICES MANUFACTURING RETAIL
Applications Market value & risk analytics Equities trading, market data Mortgage loans, antifraud Departmental applications Content management Software development & test Intranet applications Intranet infrastructure File serving, various
SAP xApps Computer aided design Departmental applications Software development & test File serving Web infrastructure, various
Internal portal, e-procurement Promotional e-mail Departmental applications Office applications POS software development File serving, fax serving Network management Web infrastructure, various
CONVENTIONAL SCENARIOS
Servers Dell, HP, IBM, various
165 servers
Fujitsu, HP, IBM
39 servers
Dell, HP, IBM, various
111 servers
Personnel 7 FTEs 1.25 FTEs 3 FTEs
VIRTUALIZED SCENARIOS
Servers POWER6 p570 8 x 3.5 GHz POWER6 p570 4 x 4.2 GHz POWER6 p570 4 x 3.5 GHz 2 x p5-550Q, 3 x p5-520Q
197 LPARs
POWER6 p570 4 x 4.2 GHz p5-520Q
44 LPARs
2 x POWER6 p570 4 x 3.5 GHz 2 x p5-550Q
122 LPARs
Personnel 2 FTEs 0.5 FTE 1.25 FTEs
As for UNIX server comparisons, two sets of scenarios were developed:
1. Conventional scenarios are based on data reported by the 21 companies, and are built around diverse bases consisting primarily of Dell, Hewlett-Packard, IBM and (for the manufacturing company) Fujitsu Intel-based servers.
Bases include different hardware technologies – ranging from Pentium 4 processor-based servers installed in 2002 to current generation quad-core Xeon processor-based servers – and different Linux distributions and versions. Conventional Linux server management and operating practices are employed.
2. Virtualized scenarios are for the same applications and workloads as conventional scenarios. As for UNIX server comparisons, it is assumed that applications are deployed on System p servers in native mode in a manner that effectively exploits the potential of System p virtualization, and that management and operating practices are supportive of this potential.
Configurations for these scenarios were developed on a case-by-case basis using LPAR-equipped System p servers to consolidate application and operating system instances.
As for UNIX server comparisons, database costs are for Oracle 10g, and calculations do not include initial license costs for copies of this. Linux systems software costs are not included.
International Technology Group 21
Cost Breakdowns
Detailed breakdowns of Linux server costs are shown in figure 21. Cost categories are the same as those employed for UNIX server comparisons.
Figure 21 Five-Year Linux Server Costs Detail
COMPANY FINANCIAL SERVICES MANUFACTURING RETAIL
CONVENTIONAL SCENARIOS ($000)
Maintenance 276.6 82.2 126.5
Database software support 210.7 235.8 89.3
Personnel 3,073.5 548.8 1,317.2
Facilities 307.6 104.7 155.4
TOTAL OPERATING COSTS 3,868.4 971.5 1,688.4
VIRTUALIZED SCENARIOS ($000)
Acquisition costs 376.3 109.2 182.2
Maintenance 112.2 32.3 31.9
Database software support 135.1 45.0 67.6
Personnel 878.1 219.5 548.8
Facilities 43.7 8.4 12.8
Operating costs (Subtotal) 1,169.1 305.2 661.1
TOTAL 1,545.4 414.4 843.3
Cost Factors
Lower five-year operating costs for virtualized relative to conventional scenarios are in most cases due to the same factors as for UNIX servers, and are summarized in figure 22.
Figure 22 Linux Server Comparisons: Factors Resulting in Lower Operating Costs
for Virtualized Scenarios
Categories Savings Factors
Maintenance 59% - 75% Fewer, newer servers reduce maintenance contract costs.
Database support 24% - 81% Fewer software copies & CPUs result in lower license, update & support costs.
Personnel 58% - 71% Fewer physical servers, reduced diversity & improved automation reduce system administration-related personnel costs.
Facilities 86% - 92% Fewer physical servers, smaller footprints, & greater energy efficiency reduce data center occupancy, power & cooling costs.
Average 67%
International Technology Group 22
Costs of Downtime
Downtime Impacts
Although IT costs tend to dominate industry discussion, the bottom-line impact of outages may be equally, if not more significant.
For example, supply chain disruptions can result in inventory shortages, missed deliveries, production and distribution bottlenecks, and other impacts. Where supply chains are tightly integrated, even brief outages can generate “cascading” effects that spread rapidly across the entire organization, and whose effects may continue to be felt long after service is restored.
Experiences with Internet commerce have shown that online outages can result in lost sales. This is not only because customers cannot make purchases at a specific time, but also because they cannot obtain information on a company’s products or services, and may look elsewhere. A customer who experiences an outage may never return, or may be more likely to divide their purchases between vendors.
Even where customer defections cannot be tied to a specific negative experience, outages contribute to overall levels of satisfaction or dissatisfaction, which affect attrition rates. These may have important bottom-line impact, particularly if lost sales or profit is measured in terms of customer lifetime value (CLV) and equivalents.
The impact on outages on productivity may also be significant. Internet and intranet applications, along with messaging and workflow systems, and a wide range of departmental and individual user tools have become deeply embedded into the business processes of many companies. Interruptions of service may be highly disruptive and, quite literally, can cause work to stop.
Effective virtualization of system resources can contribute to fewer outages. Not only risks of unplanned outages, but also planned downtime for preventative maintenance, configuration changes, software upgrades and other functions may be significantly reduced.
UNIX Server Comparisons
Costs of downtime for the financial services company represent lost operating profit, meaning net revenue after deduction of personnel, occupancy, equipment and other overheads. For the manufacturing and retail companies, the comparable metric is lost gross profit, meaning profit net of cost of goods sold (COGS), but before deduction of selling, general and administrative (SG&A) and other expenses.
Costs were determined using industry- and company-specific values assuming consistent levels of availability and cost structures over a five-year period.
For the financial services company, costs of downtime consists primarily of lost transaction fees and CLV for outages affecting online banking, brokerage and other customer self-service systems, along with CRM systems and Internet infrastructure servers supporting these.
Costs of downtime for the manufacturing company were calculated based primarily on outages affecting the company’s core SAP ERP, supply chain management (SCM) and procurement systems. For the retail company, costs were calculated based on outages affecting the company’s supply chain systems as well as Internet marketing, sales and customer service systems.
The impact of outages on the retail company’s Internet systems includes the effects not only of lost online sales, but also of storefront sales lost because customers are unable to obtain information on products, promotions, pricing, availability, store locations and other subjects.
International Technology Group 23
While online orders account for only 3 percent of total sales volume, more than 20 percent of storefront purchases are influenced by information obtained via the Internet.
Costs of downtime are greatest for the manufacturing company. This is because the company’s bottom line depends heavily upon supply chain performance, and because all of its major business systems are deployed on UNIX servers. In the financial services and retail company profiles, certain business critical systems are deployed on mainframes and other platforms that are not addressed by this report.
Linux Server Comparisons
For all Linux server scenarios, costs are for employee productivity loss. This was measured in terms of (1) idle time, reduced productivity or both for periods when applications were not available to users during working hours and (2) reduced productivity following an outage; e.g., a one-hour outage causing a 60 percent productivity reduction might be followed by a two-hour period in which a 20 percent productivity loss occurs.
Productivity loss calculations were developed for each user community supported by Linux applications in the profile companies. Costs are based on industry median salaries and benefits for large U.S. financial services, manufacturing and retail companies for the occupational groups using applications.
For example, costs of downtime for the financial services company included calculations based on median salaries and benefits for securities traders, risk analysts, mortgage loan specialists, security professionals, internal auditors and others. Manufacturing company equivalents included multiple categories of managers, analysts, engineers, technicians, salespeople and other professionals, and in some cases administrative, production and logistics personnel.
Retail company equivalents included multiple categories of managers, analysts, buyers and other professionals, along with administrative, distribution and, in some cases, in-store personnel. Software developers were included in calculations for all three companies.
Productivity losses were quantified based on the number of individuals using each application who would be affected by outages, their average remuneration per hour, and percentage values for reductions in their productivity during and subsequent to outages.
Overall costs of downtime are lowest for the financial services company. This is because Linux applications were generally more specialized, and user communities were smaller than for the manufacturing and retail companies.
International Technology Group 24
SERVER COST CALCULATIONS
General Approach
The cost and benefit implications of any IT initiative can be measured in a number of different ways. In most organizations, costs, benefits and – where this approach is employed – return on investment (ROI) values are determined on a project-by-project basis, usually over a three-year period.
While such techniques may be useful, they are inherently limited. For example, an application-specific project may deliver high levels of ROI while contributing to the fragmentation and complexity of an organization’s overall IT infrastructure. Equally, inefficiencies whose effects are minor, or at least acceptable, for small-scale deployments may have significantly larger cumulative impacts at the enterprise level.
Short-term measurements may also obscure longer-term cost drivers. Growth in server bases has often been an incremental process whose implications have not been visible on a year-by-year basis. This has made it difficult to determine baselines for efficiency measurement. Even an organization that is more efficient than it was a year or two before may still fall far short of what it could achieve.
For these reasons, it was decided for this report to base comparisons on organization-wide application portfolios, workloads and server bases, and personnel supporting these; and to focus on five-year costs.
Configuration Sizing
In translating configurations employed in conventional scenarios into System p configurations employed in virtualized scenarios, the following approach was employed.
System p configurations delivering performance equivalent to conventional scenario servers were developed for each of the main applications in profile companies. Configurations were developed for production and non-production instances.
Further calculations were then undertaken to allow for the effects of virtualization. Nominal configuration requirements were first determined for groups of instances that were deployed on separate servers, where these represented realistic candidates for consolidation onto a single System p physical server.
An overall utilization value reflecting realistic potential consolidation efficiencies was then assigned to each group of instances, and allowance was made for other factors affecting System p capacity requirements. The resulting configuration was then rounded to next largest capacity increment offered by IBM; e.g., a nominal configuration of 3.3 x 4.2 GHz cores with 12.6 GB of RAM was rounded to a 4 x 4.2 GHz System p model with 16 GB of RAM. Other hardware components were configured similarly.
Other hardware components were configured similarly. Allowance was made for other System p virtualization capabilities including use of Virtual I/O Servers and Integrated Virtual Ethernet.
Cost Components
Maintenance, software update and support, and (for virtualized scenarios) hardware and software acquisition costs were calculated based on discounted vendor U.S. list prices. For database software, Oracle 10g multi-core pricing was employed where appropriate.
International Technology Group 25
For UNIX server comparisons, systems software stacks for conventional scenarios include operating systems and, where appropriate, additional facilities providing functionality equivalent to that incorporated in AIX. Systems software stacks for virtualized scenarios include AIX and – for clustered configurations – HACMP facilities.
For Linux server comparisons, software costs are for Oracle 10g database support and update subscriptions only.
Personnel costs for UNIX and Linux system administrators were calculated using annual average salaries of U.S. $77,289 and $69,804 respectively, increased by 25.8 percent to allow for bonuses, benefits, training and other items.
Staffing levels are based on user-reported data. Definitions of system administration tasks and disciplines tend, however, to vary between organizations. The definitions employed in this report may not correspond to those employed in any individual user organization.
Facilities costs for conventional and virtualized scenarios for UNIX and Linux comparisons include costs for servers as well as power and cooling equipment.
Server facilities costs were calculated using vendor specifications for electricity consumption and footprints for server models and configurations included in scenarios. Costs include data center occupancy (calculations were based on industry standard rack mount units and service clearances for these, plus allowance for inactive areas) and electricity consumption.
Facilities costs also include hardware acquisition, maintenance, occupancy (including service clearances and inactive areas) for, and electricity consumption by power and cooling equipment. Configurations of this equipment are appropriate for overall electricity consumption and heat generation by server bases in scenarios. Cost calculations were based on U.S. specifications and discounted list prices for appropriate models from leading vendors.
Data center occupancy costs for servers and power and cooling equipment are based on a conservative assumption for annual average cost per square foot for existing facilities (i.e. costs do not include new facilities construction), while electricity costs for both are based on a conservative assumption for average price per kilowatt/hour. Both assumptions are for U.S. costs.
PSL03019-USEN-01
ABOUT THE INTERNATIONAL TECHNOLOGY GROUP
ITG sharpens your awareness of what’s happening and your competitive edge
. . . this could affect your future growth and profit prospects
The International Technology Group (ITG), established in 1983, is an independent research and management consulting firm specializing in information technology (IT) investment strategy, cost/ benefit metrics, infrastructure studies, deployment tactics, business alignment and financial analysis.
ITG was an early innovator and pioneer in developing total cost of ownership (TCO) and return on investment (ROI) processes and methodologies. In 2004, the firm received a Decade of Education Award from the Information Technology Financial Management Association (ITFMA), the leading professional association dedicated to education and advancement of financial management practices in end-user IT organizations.
The firm has undertaken more than 100 major consulting projects, released approximately 160 management reports and white papers, and delivered nearly 1,800 briefs and presentations to individual clients, user groups, industry conferences and seminars throughout the world.
Client services are designed to provide factual data and reliable documentation to assist in the decision-making process. Information provided establishes the basis for developing tactical and strategic plans. Important developments are analyzed and practical guidance is offered on the most effective ways to respond to changes that may impact or shape complex IT deployment agendas.
A broad range of services is offered, furnishing clients with the information necessary to complement their internal capabilities and resources. Customized client programs involve various combinations of the following deliverables:
Status Reports In-depth studies of important issues
Management Briefs Detailed analysis of significant developments
Management Briefings Periodic interactive meetings with management
Executive Presentations Scheduled strategic presentations for decision-makers
Email Communications Timely replies to informational requests
Telephone Consultation Immediate response to informational needs
Clients include a cross section of IT end users in the private and public sectors representing multinational corporations, industrial companies, financial institutions, service organizations, educational institutions, federal and state government agencies as well as IT system suppliers, software vendors and service firms. Federal government clients have included agencies within the Department of Defense (e.g. DISA), Department of Transportation (e.g. FAA) and Department of Treasury (e.g. US Mint).
International Technology Group
4546 El Camino Real, Suite 230 Los Altos, California 94022-1069
Telephone: (650) 949-8410 Facsimile: (650) 949-8415 Email: [email protected]
ITG
Copyright © August 2007 Gabriel Consulting Group, Inc. All Rights Reserved
The trend towards Unix virtualization is undeniable: more and more customers are embracing this new usage model to increase flexibility and lower costs. Virtualization technology allows users to safely run multiple applications and workloads on single systems - radically increasing efficiency and utilization rates while lowering overall costs. This move towards what we call “consolidated Unix” marks a sea change in the Unix usage model. The traditional Unix single workload, single system usage model led to numerous problems. First, we’re not really talking about a 1:1 relationship between workload and server. In reality, a single application (in a non-virtualized system) will require multiple servers – one each for production, development, and testing, with perhaps additional boxes for training or failover. Every new application requires several more servers, until the data center is chock full of boxes, all of which need connectivity, updating, maintenance, power/cooling, and floor space, regardless of how much or little they are actually used. It’s easy to understand why IT costs are continuing to rise, even while the cost of the base technology is continuing to fall. Expanding a bit on the topic of system utilization, customers are usually shocked by how low average utilization is on their Unix systems (generally in the area of 10-15% overall). The only thing more shocking is their x86 system average utilization, which is usually in the 5-7% range. It isn’t that these systems were purposely over-configured; it’s just that customers needed to size production systems to handle the highest anticipated peak, plus usually a 25% ‘fudge factor’ on the high side. However, most applications are fairly spiky, with deep valleys and infrequent usage peaks that are generally of short duration. This means that the average utilization for these systems is very low – even though their peaks might hit 75% or more of capacity. The utilization story for test, development, and other servers is even worse, often measured at 1-3% average utilization over time. The solution for many of the problems caused by server sprawl is virtualization – combining many applications and workloads on single systems. This technology has long been used on mainframe systems to maximize return on investment on very expensive hardware, allowing the
IBM Ups the Ante in Unix Virtualization IBM is putting pressure on their Unix rivals with the recent introduction of new POWER6 processor based systems sporting record-breaking 3.5, 4.2, or 4.7GHz processors. However, we believe that much of the value of these systems is found in the new and advanced virtualization features IBM has baked into their System p server family. This research report discusses the trend towards Unix consolidation and takes a close look at IBM’s Unix consolidation value proposition….
IBM Ups the Ante in Unix Virtualization
Copyright © August 2007 Gabriel Consulting Group, Inc. All Rights Reserved 2
systems to safely run multiple apps at high (80%+) average utilization rates. Workload management software in the operating system ensures that each workload gets the physical resources it needs to hit performance goals set by business policy. Moreover, each workload is protected so that an application failure can’t bring down the entire system or negatively impact any other application. Since this technology began in mainframe systems, it isn’t surprising that IBM is one of the first Unix vendors to aggressively add virtualization features to their Unix offerings, making them the leading Unix solution for large scale Unix server consolidation. The technology has matured considerably in the last couple of years, to the point where it is entirely safe to combine multiple applications (even mission-critical workloads) on the same system. As we stated above, customers are embracing Unix server consolidation in large numbers. Data from our last survey of Unix customers (4Q’06) shows that almost 68% of the enterprise Unix customers we surveyed have virtualized at least a portion of their Unix servers (Fig.1).
These real world customers also agree (by a 68% majority) that Unix virtualization eases system and application management (Fig.2). Having multiple applications on single systems, aided by advanced monitoring and management mechanisms present in most virtualization schemes, gives the system the ability to manage resources to ensure each application gets what it needs to handle workload spikes according to business policy – without human intervention. It also gives administrators the ability to better manage capacity needs in both the short and long term. Utilization and usage data can be captured over time, allowing administrators to easily see usage patterns and trends, and thus make better capacity planning decisions. Another advantage of virtualization, as we’re increasingly hearing from clients, is that it gives them the ability to better track actual server usage and more accurately assess chargeback and cost recovery charges. Of course, the bottom line in whether virtualization makes sense or not is, well, the bottom line. All of this higher utilization, greater flexibility, and lower cost rhetoric sounds great, but are customers really reaping significant benefits? Past GCG studies along with our most recent research shows that real world customers are indeed getting substantial cost savings from Unix consolidation (Fig.3). The level of benefit realized from this new usage model is profound enough that almost 60% of our survey respondents agree that consolidated Unix will become the norm in their organization (Fig.4). It’s interesting to note that only a very small proportion, 16%,
0%
10%
20%
30%
40%
50%
60%
"We are running multiple workloads on single Unix systems…"
% Resp. 2.33% 17.90% 12.06% 50.97% 16.73%
Strongly Disagree
Disagree Not Sure AgreeStrongly
Agree
Figure 1
0%
10%
20%
30%
40%
50%
"…Virtualized Unix systems easier to manage…"
% Resp. 3.11% 9.73% 19.46% 45.53% 22.18%
Strongly Disagree
Disagree Not Sure AgreeStrongly
Agree
Figure 2
IBM Ups the Ante in Unix Virtualization
Copyright © August 2007 Gabriel Consulting Group, Inc. All Rights Reserved 3
of participants believe that Unix virtualization/consolidation is not the way to go or the “wave of the future” in their data center.
IBM & Unix Virtualization We’ve been researching Unix server consolidation for many years now, looking at trends, customer benefits, vendor offerings, and the like. Over the last couple of years, we’ve seen a consistent pattern with customers who have standardized on IBM Unix systems – they tend to be consolidating more systems, more quickly, and with higher levels of satisfaction than customers who have standardized on other platforms. One of the reasons behind this may be IBM’s quick rise from Unix market also-ran to arguably the technology and sales leader in a very short period of time. IBM’s system sales have shot up over the past couple of years, meaning that their installed base tends to have much more of the ‘latest & greatest’ gear; in other words, systems that were designed for multiple workloads from the ground up. Another reason is that IBM has devoted a lot of resources to not just talking about virtualization or marketing consolidation, but to actually developing technology that makes consolidation safer, more efficient, and easier to implement. IBM’s recently announced POWER6 processor, along with firmware and software enhancements, significantly increases the attractiveness of System p as a consolidation platform. The suitability of a system for virtualization is dependent on not just hardware or software alone; it’s the tight integration of the various components that dictate how well the platform performs as a consolidation target. This is much of the reason why the virtualization solutions in the x86 market tend to look a little primitive when compared to what is available on Unix or mainframe systems. Virtualization vendors (such as VMware, Xen, Iron Mountain, or SWsoft) don’t control the x86 operating systems, processors, or system designs and thus can’t achieve the degree of integration that, for instance, a Unix or mainframe vendor can with their systems. The end result is that non-integrated virtualization solutions tend to have fewer features, lower automation and flexibility, and higher overhead. This doesn’t mean that they aren’t useful in many situations, but their technology certainly isn’t state-of-the art in the virtualization biz.
0%
10%
20%
30%
40%
50%
"…Virtualized Unix systems save us money..."
% Resp. 2.33% 8.17% 21.01% 45.91% 22.57%
Strongly Disagree
Disagree Not Sure AgreeStrongly
Agree
Figure 3
0%
10%
20%
30%
40%
"…Virtualized Unix will become the dominant Unix usage model in our organization..."
% Resp. 6.61% 8.95% 27.63% 34.24% 22.57%
Strongly Disagree
Disagree Not Sure Agree Strongly Agree
Figure 4
IBM Ups the Ante in Unix Virtualization
Copyright © August 2007 Gabriel Consulting Group, Inc. All Rights Reserved 4
IBM System p Virtualization: It’s the Integration IBM is pushing the virtualization ball forward on all fronts this year by introducing a new processor (POWER6), new features in their Advanced POWER Virtualization (APV), and a new version of AIX (AIX 6). The new processor, a 65 nm dual-core chip, is available in three trims: 3.5GHz, 4.2GHz, and an astounding 4.7GHz monster – the fastest production CPU in the world. What is truly remarkable about POWER6 is that while frequencies are much higher (along with performance, which is double existing POWER5 processors) it still fits into the same thermal envelope as its POWER5+ predecessor. There are a couple of ways to look at this: double the processing power for the same energy usage, or, enjoy the same processing capacity at half the energy consumption. Impressive. The first POWER6 fueled system is the p570, a midrange box with high-end capabilities. The system can handle eight dual-core processors maximum, which equals 16 cores capable of running 32 simultaneous threads. The system can also support a massive 768 GB of memory, 48 GB per core on a fully configured system. To some, this might seem like an insane amount of memory, and yes, it is definitely on the generous side. However, memory is in many cases the most important factor to consider when combining multiple applications on single systems. Most virtualized systems will become memory-bound long before they run out of CPU cycles, as virtualization mechanisms consume a significant amount of memory as they manage application and guest operating system instances. Add to this the memory that apps naturally need to function efficiently, and you find that many current systems, particularly on the low end, simply don’t have enough memory slots to provide the memory capacity needed to fully utilize the systems. This is particularly true with key enterprise workloads such as databases and scalable applications. IBM has avoided this problem by designing the system with 24 dimm slots per 2 CPU building block. The p570 is a modular system which can be scaled to an 8 processor (16 core) SMP system by way of adding 2-CPU building blocks. Each node can be directly connected to other nodes with proprietary high-speed cables. In the future, customers will be able to perform a “hot attach” (using IBM’s Hot Node Add feature) where new nodes can be brought into a functioning system without a reboot. We would expect IBM to further enhance this feature by adding the ability to perform a “hot detach” as well. The p570 can run various versions of AIX (including 5.2 and 5.3), POWER versions of SUSE Linux (now) and Red Hat Linux (soon), plus the upcoming AIX version 6, which we turn to next… AIX & Hypervisor Enhancements IBM’s newest version of their Unix AIX operating system, AIX 6, is expected to ship sometime in November. In a bit of a departure from previous o/s releases, they have made AIX 6 available as an open beta, beginning in July. From a feature and functionality perspective, IBM has done a solid job of pushing AIX forward on a number of fronts, including availability, manageability, and security. Almost all of these enhancements have at least some impact on the suitability of the platform as a consolidation target. However, the list of enhancements is fairly long, and this report is about virtualization, so in the interest of brevity, we’re going to drill down into a couple of important new features that most directly relate how the platform will perform as a consolidation host.
IBM Ups the Ante in Unix Virtualization
Copyright © August 2007 Gabriel Consulting Group, Inc. All Rights Reserved 5
The first of these is what IBM is calling AIX Workload Partitions (WPARs), which adds a new way to combine workloads onto a single system. Currently, IBM customers generally rely on the LPAR (Logical Partitions) mechanism to run multiple workloads on a single system. Each LPAR has its own unique instance of AIX (or Linux) and is given a share of physical system resources to use. LPARs are logically isolated from each other, and a problem in one LPAR can not negatively impact workloads running in other LPARs. Customers consolidating on System p systems set up individual LPARs for each workload or application hosted on the system. WPARs are a different, but complimentary, way to get to the same end result. WPARs manage multiple applications within a single AIX operating system instance. Each application or workload is given shares of the resources assigned to the WPAR and has a high degree of logical isolation from other workloads in the same WPAR. These shares can be dynamically (and automatically) adjusted to quickly change with changing business requirements. Important workloads receive priority for the resources they need to meet SLAs. Less critical workloads can be dynamically throttled down so that more resource can be used to satisfy critical application peaks. With WPARs, customers can reduce the number of operating system instances they need to manage while also reducing virtualization processing overhead. Hundreds of WPARs can be run on a single instance of AIX 6 and WPARs can even be placed within LPARs for maximum flexibility. IBM has also developed a WPAR manager (cunningly named “Workload Partitions Manager for AIX”) that will allow customers to manage and monitor WPARs across multiple physical systems from a single GUI-based console. WPARs are an important new capability that helps to complete the System p virtualization story. In order to get maximum system utilization and flexibility, customers need both LPAR and WPAR virtualization mechanisms. From a processing standpoint, WPARs are more efficient – more useful production per cycle due to the amount of resources saved by utilizing a single instance of the operating system. However, not all applications in an enterprise run on the same version of a particular operating system – this is where having separate LPARs helps get the most out of an individual server. LPARs, with their individual operating system instances, provide a higher level of workload isolation than WPARs. Modern data centers who need to get the most out of their infrastructure investment will find themselves using both LPARs and WPARs. The WPAR feature is included in the base version of AIX 6, along with single system management capabilities. AIX 6 will run on POWER6, POWER5, and POWER4 based systems. Another new feature in AIX 6 is Live Application Mobility, which, as its name implies, allows users to move applications inside of WPARs from one physical system to another while the application is active. This feature can be used to balance workloads across systems, to depopulate systems so that they can be taken down for maintenance, or to reduce data center energy usage. Live Application Mobility freezes the application and WPAR state in time, and then transfers the state information to the target system, where it is fired up again. The actual transfer process is reasonably quick, taking from a few seconds to a few minutes, depending on the network speed and the size of the workload being transferred. While we’re talking about moving workloads around, this is a good time to discuss another important new capability: Live Partition Mobility. This feature will be included with the upcoming version of IBM’s APV (Advanced POWER Virtualization) hypervisor and gives customers the ability to move live LPARs (and the workloads hosted within them) to another
IBM Ups the Ante in Unix Virtualization
Copyright © August 2007 Gabriel Consulting Group, Inc. All Rights Reserved 6
physical server – without losing in-flight transactions and without any workload interruption. Here’s how it works: when the transfer is launched, a new LPAR is created on the target system and memory data from the source system is rapidly copied from source to target. The source LPAR continues to operate normally, but all the while, the target LPAR continues to copy data – getting closer and closer to being a clone of the source LPAR. When the difference between source and target is minimal, the switchover is triggered – including memory, in-flight transactions, I/O contents, disk ownership, etc. When this final switchover takes place, users will see, worst case, a 2-second pause in the application. Two seconds is less than a ping timeout, which is why apps will not be interrupted. The process is transparent to most applications and middleware, but can take some time to accomplish – IBM estimates it will take 10 minutes to move an LPAR with 64GB of memory, and maybe half an hour for an LPAR with 144GB of memory. But the time the process takes to complete is irrelevant in the vast majority of cases due to the fact that application processing goes on as usual – same speed, same throughput, throughout the transfer process. With Live Application Mobility and Live Partition Mobility, IBM has made a solid move towards eliminating planned downtime – at least from an application standpoint. Customers will be able to quickly, safely, and easily move workloads from systems that, for example, need to be taken down for maintenance, and place them onto other, perhaps under-utilized servers. Since planned downtime might account for as much as half of all downtime, these features are indeed very significant advances and will give IBM a leg up on their competitors who have not yet announced their plans along these lines. It will also give cutting-edge customers the ability to more closely manage their energy consumption. Highly utilized systems (35%+ CPU utilization) are much more energy efficient, on a workload per watt basis, than systems running at lesser utilization rates. Customers who opportunistically move workloads to raise system utilization will then be able to power down idle boxes, thus saving the energy consumed by the shut down boxes plus lowering overall cooling demands. New Virtualization Frontiers with pAVE IBM has expanded their virtualization market with the development of their System p Application Virtual Environment (pAVE) that allows most native x86 Linux applications to run un-altered on System p hardware. The pAVE technology translates Linux system calls so that they can be executed on POWER processor-based systems. Linux applications can be run in LPARs utilizing Red Hat 4 (update 5 for POWER6) or Novell SLES 10 (service pack 1 for POWER6) operating systems. From an operational standpoint, there is little or no intervention necessary – the system ‘just knows’ (as IBM puts it) when the application is a native x86 Linux binary and runs it in pAVE mode. There isn’t a lot of performance data this early in the game. There’s certainly going to be some additional processing overhead, but there are at least two factors that may reduce negative impact. The first is that the new POWER systems are much faster (and have more memory) than current x86 systems. The second is that code that executes in pAVE mode is cached in a way so that subsequent runs of the same code will execute faster. However, the potential benefits from pAVE aren’t performance based – at least not in the way that we traditionally talk about application performance. The business value proposition for running native Linux code on System p boxes has much more to do with the efficiencies arising from virtualization than with raw performance. The IBM hypervisor, combined with System p
IBM Ups the Ante in Unix Virtualization
Copyright © August 2007 Gabriel Consulting Group, Inc. All Rights Reserved 7
Advanced POWER Virtualization (APV), offers much more robust virtualization features and capabilities than the current crop of x86 virtualization schemes. For example, with IBM’s APV, partitions can be as large as the entire system, while the leading x86 virtualization product is limited to four processors. Also, under APV, partitions can be dynamically created, deleted, or changed without requiring a reboot of other partitions or the entire system. For System p data centers, pAVE will allow them to radically improve data center efficiency by consolidating Linux applications onto to System p hardware. Average x86 system utilization hovers around 6%, which means there are many consolidation opportunities in most IT shops. For IBM, pAVE gives the opportunity to capture more workloads on System p hardware and perhaps persuade more ISVs to recompile their code so that it runs natively on POWER versions of Linux. Summary & Recommendations Over the past several years, it’s become clear that the era of speeds & feeds is over. While IBM’s new systems will get plenty of attention due to the 4GHz+ engine under the hood, we believe that other features and factors will provide the higher business value to customers who need to get the most out of their server dollar. The battleground of the future is how well the systems perform as corporate assets. By this we mean factors such as the useful throughput a system can provide on real customer workloads, the utilization rate of the system, and how well it minimizes facilities and labor costs. This is a whole new measure of ‘bang for the buck’ and takes in much more than benchmark scores and traditional hardware/software cost-per-transaction. The vendors who will succeed in this era are those that concentrate on providing overall business value, not in exhibiting technology for technology’s sake, but in making sure their products provide business benefits over and above competitive offerings. In our view, IBM has raised the business value bar significantly with the new POWER6 offerings – not because of raw speed, but because of the sophisticated way the systems can be used to provide business value and earn their keep as business assets.
Entire contents © 2007 Gabriel Consulting Group, Inc. All rights reserved. This document may not be reproduced or transmitted in any form by any means without prior written permission from the publisher. All trademarks and registered trademarks of the products and corporations mentioned are the property of the respective holders. The information contained in this publication has been obtained from sources believed to be reliable. Gabriel Consulting Group does not warrant the completeness, accuracy, or adequacy of this report and bears no liability for errors, omissions, inadequacies, or interpretations of the information contained herein. Opinions reflect the judgment of Gabriel Consulting Group at the time of publication and are subject to change without notice.
phone / 503.372.9389
[email protected] www.gabrielconsultinggroup.com