Blog Central

Keep up to date...
Read blogs submitted by our partners, customers, and vendors.

• RoHS WEEE Compliance
• Jonathan Schwart's
  Weblog
• Network Appliance, Inc.
  Dave's Blog
• The Console / VMware Blog

Improving Data Center Efficiency by Thinking Green

Volume 1, 2007

Although an increasing proportion of today's information is virtual, the cost of collecting, storing, processing, and reporting that information is very real. Building a data center and supplying it with hardware, software, and a support staff is certainly expensive. But the electricity and cooling required for these computers may well dwarf other expenses.

This situation will only get worse with time. As the desire for information increases, so does the need for computing power. That means more hardware to house, power, and cool.

Many businesses have begun to explore alternatives to reduce their impact on the world and their neighborhood. The good news is that vigorous research is underway to improve the efficiency of computing in general, from individual machines and processors all the way to large data centers. But what can companies do in the meantime?

Physical Limitations

It's difficult to ignore the pervasive trend toward horizontal scalability through commodity computing. Rather than buy the single largest, most powerful computer available, companies often purchase several identical, much cheaper computers and divide the workload among them. There are many advantages, from price to redundancy.

This is a useful technique used by industry-leading companies like Google.

There are also many drawbacks. Instead of one machine, there are now racks full of machines drawing power and producing heat. Electrical costs increase, hardware failure rates go up, and the moisture levels in the data center rise to dangerous levels. There are no easy or cheap solutions to counter these problems. Building a new data center is expensive, even one in a suitable location with sufficient power. Replacing all or most of the machines is cost-prohibitive as well. In either case, you're just fighting to stay ahead of the problem, since it's unlikely that demand for accurate, timely data will decrease.

Better Processors

Most of the electricity used and heat generated inside a computer comes from its CPU. Imagine a room full of 100-watt light bulbs running at full power all day every day. That's what a server room is like. In computing, heat not only represents wasted energy, but is also a threat to the equipment. Removing heat from the server room only adds to the energy waste.

Businesses that deploy one application per physical machine (a common scenario) are typically running computers at 40% of their maximum capacity. While this practice leaves room for growth, the reality is that each computer is using the same amount of energy, whether it's using 40% or 100% of its capacity. That is grossly efficient.

Newer processors are starting to do more with less. Chip manufacturers have experimented with lower voltages and now sell chips that perform as well as their counterparts while drawing less power and producing less heat. A similar approach is to throttle the CPU while it runs at less than full capacity. Some CPUs can disable entire features when they're not in use; the fewer transistors running, the less electricity needed.

Perhaps the future of CPUs also lies in horizontal scaling. Rather than increase brute power, many new CPUs have multiple cores. Think of them as multiple CPUs all on a single chip. Though individually they may be smaller and slower than one huge monster CPU, they perform multiple tasks simultaneously. This divide-and-conquer approach can produce the same results far faster and less expensively.

Better Cases

Heat is the natural enemy of computing: it increases defect rates and often contributes to dangerous humidity levels. A great deal of hardware design focuses on how to remove heat from the inside of the case. This is particularly important for server equipment that has multiple CPUs and stands in racks alongside other computers banked in rows.

Inefficiencies in heat dissipation can be deadly to machines.

One of the biggest offenders is the bog-standard power supply, which may operate at only 80% efficiency. That 20% wasted power becomes heat, which requires fans inside and outside of the case. Fans, in turn, draw their own electricity and produce their own heat. Improving the efficiency of power supplies to 90% reduces the amount of waste heat generated and the necessary draw by half. Further improvements, coupled with lower-voltage CPUs, will produce further gains.

Even the design of computer cases themselves can make a substantial difference in power requirements. The optimal airflow through a computer will vent heated air without leaving hot spots—places of likely failure—within the case. IBM's Blue Gene/L supercomputer uses a unique non-cube shape, designed explicitly to move heat through the case away from racks and racks of stacked CPUs.

Rack Design

Blue Gene's design works with basic principles of physics. Those same principles explain why machines at the top of a rack tend to have higher failure rates than machines at the bottom of a rack: heat rises.

Cooling inside the case relies on having a steady supply of cooler air outside the case. If none is available, due to a dozen other machines underneath or in front providing superheated air, the chances of cooling your CPU diminish. Your fans will spin, but they'll have little effect.

More sophisticated rack hardware can mitigate this tendency. A heat exchanger on the rear door can draw heat away from the rack altogether. Using water or other liquids for cooling can be several times as efficient as air. Applying this technique to individual racks may be less expensive and time consuming than retrofitting an entire facility.

Server Room Redesign

For companies that have the budget to go beyond incremental improvements, rethinking the entire data center may provide further efficiency gains. A thorough profile of available space can reveal thermal differences at various points in the room. Airflow analysis is critical to cooling, but guessing at temperature gradients in a three-dimensional space is very difficult without accurate measurements. Moving one or more cool air vent or hot air return can increase efficiency dramatically.

Software Changes

Even gradual improvement of hardware can take time and, as usual, as much money as a business has to spend. Other options require software or configuration changes, but can dramatically slow the growth rate of new hardware. One cost-effective alternative for increasing the efficiency of existing hardware is virtualization.

Rather than deploy a new box for every new application, it's possible to deploy a virtual machine image on existing hardware. Where the dedicated box may use 40 to 50% of the machine's capabilities at most—while requiring all of its space and energy overhead—even just two virtual images running on a single box can cut machine requirements in half without adversely affecting performance.

In theory, you can work to the full capacity of your existing machines. In practice, you still may want a little bit of overhead—but you can manage that breathing room on a center-wide basis.

Virtualization also provides several other benefits, especially related to ease of administration. Deploying a new image can be as easy as launching an application. Migrating an image to a dedicated box (or a less-populated box) may be almost as simple.

Conclusions

Going green in a server room may be an important goal for ethical reasons, but it can also improve the bottom line. Improved efficiency can reduce costs and slow the rate of data center growth. Potential solutions include ideas as simple as revising HVAC systems, and as complex as redesigning the thermal profile of machines at the transistor level.

Though you may see most improvement from changing hardware and physical layouts, the number of factors involved is large, and their interactions are complex. Proper calculations will take into account not only mathematics and physics, but the costs and tradeoffs, as well as your proposed growth rate.

Fortunately, research continues at a brisk pace to reduce the power consumed and heat produced by computers. The industry is reinventing itself yet again to meet these new challenges. Like many of the large problems in technology, there's no single answer and no single organization capable of providing all of the answers. As IBM's Jim Gleason puts it, "The problem's going to get solved collaboratively, somehow."

Recommend this article Subscribe RSS Feed Print