Innovating for the environment:

Traditional systems measure data center temperature at the hot-air return of air conditioning units instead of at the heat source – the racks. Because controls are imprecise, these data centers are often operated at less than maximum capacity to avoid overheating. That's an expensive proposition.

Researchers developed a distributed sensor network attached to standard racks that provides a direct measurement of the environment where it is most useful. The technology then processes sensor data to determine how best to allocate cooling resources to maintain specified rack temperatures.

Much of this work has been incorporated into HP's Dynamic Smart Cooling technology. Environmental data collected from sensors is used to automatically adjust the provisioning of cooling resources using standard data center environmental control equipment. This technology has the potential to cut data center cooling costs by 25 to 40 percent.

In addition to saving energy, Dynamic Smart Cooling makes the data center's cooling infrastructure more efficient. This allows increased computing capacity while improving the reliability of environmental control system.

Energy consumption and thermal requirements are fast becoming the limiting factors in data center environments. Increasing server densities and workloads pose power and cooling demands that could require data center operators to renovate existing facilities or even build new ones.

Data Center Thermal Assessment Services use sophisticated modeling techniques pioneered at HP Labs to provide detailed reports into the data center’s thermal footprint, make comparisons to best practices and deliver recommendations about how to improve energy efficiency.

The assessments can also be used to determine the best design for new data center environments.

The services include a capability called Thermal Zone Mapping, an analysis and visualization tool that uses data generated from computer models to produce a three-dimensional model of how much and where data center air conditioners are cooling. This allows customers to arrange and manage air conditioning for optimal cooling, increased energy efficiency and lower costs.

When it comes to cooling, not all locations in the data center are created equal. Some racks are easier to cool (perhaps they are closer to the air-conditioning units) while others (perhaps those in a corner) are more difficult to cool. At the same time, different servers generate more heat, depending on their design and computational workload.

The result: hot spots in the data center, which require the air conditioning system to work harder  simply to ensure that no server gets too hot.

What if you could allocate computing workloads to the most thermally efficient locations in the data center? That is the proposition behind cool job allocation, a method of monitoring and ranking server locations according to cooling efficiency. This information could then be used to determine where to place compute jobs and provide guidance for workload consolidation.
 
In early experiments, researchers found that allocating compute workloads based on environmental factors reduced energy used to cool servers by 30 percent when compared with random workload placement.

Power delivery, electricity consumption and heat management are key challenges in data center environments. Although there is a variety of solutions to these problems, each operates in isolation – regulating power using different techniques at different levels of the hardware and software stack – and there is no way to coordinate them all.

As a result, these solutions are likely to interfere with each other in ways that may decrease their effectiveness, or even lead to incorrect behavior.

Working with HP business groups, researchers are trying to solve this problem by creating a unified power-management architecture as well as performing analysis that provides insight into different architectural trade-offs. One such example is ensemble-power management that does power capping across a collection of servers.

Our goal is to unify this into an end-to-end power-management system that incorporates performance, cooling, and power management -- using models to track and measure heat flow and performance throughout the data center – and provides insights into the benefits and drawbacks of different data center design options.

As global IT power consumption becomes a pressing concern, IT providers are scrambling to create or configure systems for maximum energy efficiency. But the only way to determine true energy efficiency is to have a common benchmark that takes into account not just power savings, but also system performance. For example, one system configuration may offer the most power savings, but it may run much more slowly and therefore use more total energy.

The common benchmark should also take into account the power used by the whole system rather than just its individual components. A benchmark focusing only on CPU, for example, would fail to account for power used by other components such as disks and memory, which comprise a significant portion of the total power used.

With a team from Stanford University, HP researchers developed JouleSort (named for the Joule, a standard unit of energy), an extension of the Sort Benchmarks that are traditionally used to measure performance/cost performance of computer systems.

JouleSort measures system performance when sorting a fixed, but large amount of data. This workload exercises all core components of a system and is portable -- meaning it is capable of evaluating the energy efficiency of a wide range of computer systems. JouleSort helps system designers understand the bottlenecks in achieving energy efficiency and points out trends in energy-efficient designs.

The team applied this benchmark to create a new system that was nearly four times more energy efficient than the previous estimated best system from 2006. This work was first presented at a conference in China in June 2007.

Since then, JouleSort has been adopted by the database community’s Sort Committee that administers other sort benchmarks for pure performance and cost.

Longer term, researchers are exploring a way to measure the environmental impact of the entire IT lifecycle – from the raw materials extracted to build the machine to its manufacturing to its recycling and potential reuse.