Measurement and Management Technologies for Data Centers - overview
The MMT (Measurement and Management Technologies) work was initially motivated by surging
energy consumption in data centers (DCs). DCs form the backbone of the information
technology (IT) infrastructure with literally every business and government entity
critically depending on it. They house very large assemblies of densely packed IT equipment
and can be easily as large as several football fields on multiple floors, with a total power
consumption of tens of mega-watts, which is a significant fraction of a full power plant.
While today with 2 % of the total US energy consumption the impact of DCs is relatively
minor, the rate of increase (~15 % per year or doubling every 5 years) is alarming. DCs are very complex systems, with multiple applications running on various servers,
distributed throughout the facilities with different network, storage, processing, and
memory requirements. The challenge is that DC energy management encompasses very different
technical domains, ranging from facilities, power delivery, and IT systems to the actual
core technologies. Traditionally, the management the IT infrastructure relies on a high
level of abstraction, which on one hand makes it easier for the administrator to deal with
the complexity but in essence hides information about which physical locations the different
applications might be running in. However, in order to reduce energy consumption, the
physical and logical layers have to be managed holistically. To improve energy efficiency in DCs we have developed MMT (Measurement and Management
Technologies) that allows simultaneous measurement and management of both the logical and
physical IT infrastructure. First, we developed a comprehensive measurement system, which
included wireless mesh networks as well data adapters to collect energy information from the
IT systems and the support infrastructure. The main innovation of the technical work is
associated with the development of an accurate and fast model that allows operators and
control systems to reduce energy consumption. Traditional approaches, to model for example
heat distributions, relied on complex computational fluid dynamic models, which are not only
slow but also computationally very intensive when used for operational control. We developed
several physics-based models, which model much faster with similar or better accuracy. In
essence, these models leverage the real-time measurement data, which is applied as boundary
nodes and thus enable drastic simplification of the complexity of the physical description.
We complemented these physics models with advanced spatial kriging models to further improve
the time response. The work includes the development a series of control and optimization technologies, which
range from decision-support systems using advanced visualization technologies, closed-loop
optimal control system to full scale optimization. As part of this effort, we developed an
intelligent MMT software platform, which provides the underlying basis for these control
technologies. For example, the technology offers real-time energy efficiency reporting
features. Chilled air pressure is constantly monitored and displayed, which gauges air
leakage as well as service disruptions in the cooling system. It allows managing cooling
power supplied to the different pieces of IT equipment and the rendering of real-time
temperature distributions. Optimization algorithms can determine the optimum placement of
workloads based on the current cooling configuration, or a series of automatic controls for
the physical infrastructure (such as air conditioning units, chiller system, power
distribution units, etc) can be used to support the IT load with optimum energy
efficiency. Today almost 500 data center facilities are using MMT with approximately 50% of users in the
Americas, 40% in Asia-Pacific, and 20% in Europe and Africa. MMT has achieved, on average,
more than 10% of energy efficiency improvements resulting to date in annual energy savings
of more than 500 Million kW-hours, which is a sizeable fraction of a large scale power
plant. Beyond cost savings (more than $30M annually to date) the technology has avoided
major capital investments, which can be estimated to be more than $150M. Demonstration and
showcases of this technology have been installed in Southbury (CT), Poughkeepsie (NY), and
Raleigh (NC) as well as overseas in Boeblingen (Germany), Montpellier (France), Dublin
(Ireland), Delhi (India), and Shanghai (China) to help educate customers in the
implementation of more energy-efficient and sustainable IT practices. One of the virtues of the technology lies in its universality. MMT has expanded
telecommunication facilities and network switching offices, working with ATT, as well as
buildings in general. We went even beyond that and made use of MMT at the NY Metropolitan
Museum of Art, where this technology is being used to monitor and manage the preservation of
objects of art.
One
Two Three