|
|
|
|
IT & DATA MANAGEMENT RESEARCH,
INDUSTRY ANALYSIS & CONSULTING
Introduction
Enterprise Management Associates (EMA) devotes considerable time to discussions of current and
desired IT management technologies with CIOs and other IT operations personnel. Many consider an
ideal management stack to include the following features and capabilities:
• Top-down modeling of business applications and dependencies
• Holistic, policy-based, vendor-agnostic management of compute, storage, and network elements
• Virtualization-aware management, including network, compute, storage and virtual machine
resources
• Definition of business- and element-level policies, including application prioritization, required
performance levels, and desired optimal states
• “Stateless” management capabilities that abstract business applications from underlying hardware
• Single management console that incorporates all management capabilities while requiring a single
multi-purpose agent on managed devices
• Orchestration capabilities that fluidly combine all management capabilities, increasing data center
efficiencies while reducing overhead, saving money and resources
This wish list is a tall order to fill, but certain vendors are making excellent progress toward fulfilling
these requirements. Effective IT management provides many or all of these benefits:
• Maximization of data center resource cost/benefit ratios
• Increased application availability and performance
• Decreased costs from automation of routine IT tasks and decreased hardware requirements
through reduced spare requirements
• Increased compliance with corporate, IT and governmental regulations and policies
• Decreased power requirements; increased data center scalability and density
• Rapid adaptability to changing business demands, including provisioning of additional resources
as needed to satisfy “burst” demands
This paper will focus on the application and benefits derived from this new breed of holistic IT
management technologies, with an emphasis on increased efficiencies, reduced costs, and the environ-
mental benefits that result from them. It then provides an in-depth comparison of the management
and server offerings from two industry heavyweights, Cisco and HP, and concludes with an EMA
analysis of these trends.
Business Drivers toward a New Management Paradigm
The drive to increase data center efficiencies has been unrelenting for a decade or more, and IT man-
agement is a key focus for many organizations seeking to reduce costs and minimize the environmental
impact from IT operations. This trend shows no signs of slowing down as organizations grow ever
more dependent on IT every day. Pressure to decrease power consumption and increase data center
density are seemingly at odds with increasing demand for compute, storage and networking capabilities.
At the center of the storm is the increase in server deployments, which EMA estimates will continue to
increase at a minimum of 10% CAGR for the next few years, incrementally accelerating over time.
Data Center Management: The Key Ingredient for Reducing Server Power while Increasing Data Center Capacity
©2010 Enterprise Management Associates, Inc. All Rights Reserved. | www.enterprisemanagement.com
Page 1
IT & DATA MANAGEMENT RESEARCH,
INDUSTRY ANALYSIS & CONSULTING
According to the U.S. Environmental Protection Agency (EPA), electricity consumed by servers in
U.S. data centers in 2006 represented about 1.5% of national electricity use, or around 61 billion
kilowatt-hours (kWh)—equivalent to approximately 5.8 million average U.S. households1 and result-
ing in energy costs of approximately $5.4 billion while emitting 36 million tons of carbon dioxide
into the atmosphere.2
Power consumption has emerged as a primary focus of many organizations that have undertaken
“Green IT” initiatives. While reducing an organization’s contributions to greenhouse gas emissions
is compelling and necessary from a corporate responsibility standpoint, an even larger impact on the
bottom line is created when a data center runs out of power. While every watt is important, the “last
watt +1,” which occurs when a data center runs out of power capacity and necessitates the build-out
or relocation of the data center, costs tens or hundreds of millions of dollars. This scenario creates a
massive stepping function and executive review, requiring data center managers to show that they have
squeezed every watt out of the company’s existing data center investment. Data center scalability has
therefore emerged as a key driver in the “greening” of the data center. Organizations are discovering
that data centers designed to provide the highest level of performance per square meter at the lowest
power levels, can derive up to 30% additional capacity from existing facilities. They also reap the PR
and cost reduction benefits of a green data center in the process. Proactive planning today can elimi-
nate or delay the necessity of expanding or relocating the data centers of tomorrow, saving millions of
dollars in the process.
“Increasing power density can lead to a situation in which companies are forced to build new data centers not because
they are running out of floor space but because they need power and cooling beyond what can be provided in their
existing data centers. This situation has driven much of the recent interest in energy-efficiency improvements for data
centers. If the power consumed (and resulting heat generated) in data centers can be reduced through energy-efficiency
measures, the existing infrastructure can continue to meet cooling and power needs, and costly investments in new
data centers can be deferred.”
Report to Congress on Server and Data Center Energy Efficiency—Public Law 109-431, United
States Environmental Protection Agency ENERGY STAR PROGRAM, August 2, 2007
There are two fundamental ways to reduce data center power requirements: 1) purchase energy efficient
equipment; and/or 2) employ “power-friendly” IT management technologies. The latter can reduce
power consumption on a much larger scale than simple hardware replacements. This is true only if the
data center management stack is not so complex that it causes management overhead to spiral out of
control, increasing expenditures more (and faster) than the power savings realized.
The Automated, Adaptive Data Center
Vendors created great hyperbole over the past decade, hailing the advent of a fully automated, adap-
tive, on-demand data center. In this brave new world, data center elements are grouped into generic
resource pools while an all-knowing, all-seeing management genie, sometimes known as an orchestra-
tor, manages the entire data center based on business policies. The “orchestrator” ensures that all loads
are matched with the appropriate amount of with resources, delivering maximum ROI and business
performance at the lowest cost.
Data Center Management: The Key Ingredient for Reducing Server Power while Increasing Data Center Capacity
©2010 Enterprise Management Associates, Inc. All Rights Reserved. | www.enterprisemanagement.com
Page 2
“Report to Congress on Server and Data Center Energy Efficiency—Public Law 109-431”, United States Environmental
Protection Agency ENERGY STAR PROGRAM, August 2, 2007
UK National Energy Foundation Energy to Carbon Converter: http://www.nef.org.uk/greencompany/co2calculator.htm
1
2
IT & DATA MANAGEMENT RESEARCH,
INDUSTRY ANALYSIS & CONSULTING
Unfortunately, the grand vision outlined above never progressed beyond the vaporware stage due to a
lack of control at the element level. Passive, unintelligent hardware controlled by an elegant orchestra-
tor is akin to a grade school band led by a great director—except that dumb hardware never learns
or improves! The primary gap between vision and reality has been hardware that lacked the ability to
perform autonomous management actions based on business policies set by a centralized management
engine. It is no longer acceptable to have a centralized orchestrator that attempts to manage dumb ele-
ments directly—the elements must themselves be intelligent and self-managing, dynamically adapting
to changes without requiring the overhead of an orchestrator.
The advances of virtualization technologies is a key technical enabler toward these goals, allowing
intelligent network, storage and compute resources to be pooled together and managed as a group.
Provisioning technologies provide a level of automation, quickly assigning virtual resources to business
applications when needed.
The biggest virtualization development to date is the Virtual Machine, or VM. VMs encapsulate an
entire business application environment inside of a virtual, portable “bubble” that can be easily moved
from server to server (or even into the Cloud) as needed. Combine a VM with virtual compute, net-
work, and storage capabilities and an automated provisioning system, and truly portable business ser-
vices started to become real (though still with strategic gaps). The VM is becoming the “atomic unit”
of the data center, around which virtual resources are deployed as needed. It is important to realize,
however, that VMs are not suitable for all purposes, and monolithic deployments like Oracle, with large
memory footprints, will continue to exist for many years to come. Technology vendors need to be able
to support fully virtualized, partially virtualized, and non-virtualized environments.
Until recently, flaws in virtualization management technologies prevented the fully automated data
center vision from becoming a reality. First, many so-called virtualized resources still require a large
amount of manual tweaking during the provisioning process in order to function correctly. For
example, many “virtual” networks still require specific configuration settings be applied for certain
applications or VMs, for certain types of “virtual” hardware. This is because some vendors have not
delivered a hardware manager that fully completed the work required to package all of an application’s
configuration items, policies and dependencies into a fully portable, policy-based package. Without this
fully portable package, there is still a manual intervention requirement to make all of the elements work
together every time an application is moved or additional resources are provisioned. This capability is
required across all server/network/storage environments, whether virtualized or not.
Once a stable hardware management base is attained (which includes deployment of intelligent, self-
managing, policy-aware elements), an overarching, policy-based manager (the genie, or director) can be
implemented. The bi-directional integration of the manager with the director is a critical component,
and maximum flexibility/scalability by the hardware manager is key.
The director’s job is to serve as a centralized repository for business and IT policies, ensuring that busi-
ness-critical applications receive the resources they require in order to enable the business. IT and the
business work together to define policies that govern all aspects of a business services, ranging from
the type and quantity of resources required, to rules controlling the provisioning of extra capacity to
satisfy bursts in business demand, to policies ensuring security.
The director, thanks to an automated, policy based hardware manager, can even mandate movement
of VMs and their associated applications from server to server in situ (via the hardware manager)—without
Data Center Management: The Key Ingredient for Reducing Server Power while Increasing Data Center Capacity
©2010 Enterprise Management Associates, Inc. All Rights Reserved. | www.enterprisemanagement.com
Page 3
IT & DATA MANAGEMENT RESEARCH,
INDUSTRY ANALYSIS & CONSULTING
interrupting running business processes and security protocols and policies. This process of con-
tinuous resource optimization yields the highest levels of data center performance while maximizing
resource utilization and security.
Maximizing Data Center Performance While Reducing
Cost through Automation
Historically, lack of management and virtualization maturity forced IT organizations to over-provision
resources to ensure that adequate spare capacity existed to satisfy cyclical and unforeseen demand
spikes, provide disaster recovery and meet failover requirements. Common practices included allocat-
ing an extra backup server for every critical production server in the data center, some even going so
far as to reserve two extra servers in order to ensure triple redundancy. Bare-metal server provisioning
was impractical, requiring too much time and manual intervention to make it viable, which required
fully provisioned backup servers to be running at all times. These practices are an incredible waste
of human and system resources in addition to being very inefficient from a footprint and power
perspective. Even though backup servers consume a fraction of the power, hundreds or thousands of
idle servers collectively require a large amount of power and significant amounts of non-production
oriented data center space.
In an environment with a highly effective hardware manager as described above, a highly automated
data center (independent of virtualization) pools spare servers in a bare metal state, making them
holistically available for provisioning to virtually any task within minutes. A large majority of spares
are powered down, since they can be activated and provisioned from bare-metal state in minutes when
needed. This reduces the total number of backup servers required since servers are now consumed on-
demand, quickly allocated and de-allocated as needed, shared by multiple business services. This saves
a large amount of the power formerly consumed by active backup servers. Bare-metal provisioners
must work equally well for VMs as well as “classic” (monolithic or non-virtualized) server data center
architectures.
The advent of blade computing several years ago is now reaching widespread acceptance, particularly
due to significant advantages from power, efficiency and management perspectives. Blades pack a lot
of compute power inside of a small, modular physical space, providing a generic compute resource
that can be quickly and easily “hot swapped” if it fails. Couple these hardware advantages with the new
breed of automated management technology that allows rapid bare-metal provisioning, policy-based
virtualization and an orchestrator that dynamically and automatically moves workloads when a blade
fails, or if business demand changes, and blade computing can provide a very strong business case.
Data Center Power Considerations
As mentioned previously, power capacity is a key data center and cost efficiency driver. In addition
to the core power requirements for data center elements, every dollar spent powering the data center
also incurs a dollar cost to power, heat and cool the hardware. As data center densities continue to
increase, power constraints often limit scalability long before physical space runs out. Maximizing
power utilization efficiencies not only increases data center capabilities, but it also decreases power/
performance cost and environmental impact. As discussed above, the real business benefit is increased
compute carrying capacity of existing data centers, deferring or completely avoiding additional data
center build-outs.
Data Center Management: The Key Ingredient for Reducing Server Power while Increasing Data Center Capacity
©2010 Enterprise Management Associates, Inc. All Rights Reserved. | www.enterprisemanagement.com
Page 4
|
|
|
|
