This piece was originally published in the July 2016 issue of electroindustry.
John Caskey, Vice President of Operations, NEMA
The three most important factors in selecting real estate are location, location, and location—although size, craftsmanship, and other features may also be important. When it comes to data centers, the determining factor is reliability.
While data centers have been around for decades, the bar has been raised on what constitutes a state-of-the-art facility. Some notable changes have led to the following:
- Leased space (rather than owned facility)
- Dynamic energy storage (rather than static battery storage)
- Meticulously maintained diesel generators
- Higher-temperature server rooms
- Interconnected systems using a ring bus
Before the explosive growth of the internet, many data centers were designed, built, and maintained by the entity that used the data: a government agency, insurance company, bank, university, or other organization.
As the internet expanded, many businesses decided to lease space in data centers owned and maintained by other companies. The businesses that collected and used the data relied on other companies to construct the buildings and design the electrical infrastructure to ensure high electrical reliability.
The leasing of many data centers has led to the development of numerous real estate investment trusts (REITs). This creates a win-win scenario: REIT investors are happy because they get an almost-guaranteed return on their investments; data users are happy because they can lease state-of-the-art data centers with 100-percent reliability without having to make significant investments in the building structure. Some data-center users, however, still build their own centers or lease from non-REIT companies.
Doubling Down on Redundancy
The conventional data center uses electrical service from the local utility, often provided through underground cables. It then backs up that power with additional cables, transformers, and switches connected to uninterruptable power supplies (UPSs), battery storage, and diesel generators. Many data centers still use this relatively simple approach with success. Other companies, however, enhance the design by adding greater redundancy to ensure that the power never goes down.
Several years ago, 100 percent of online energy storage was in flooded lead acid batteries. These batteries were stored in isolated rooms with special systems to ensure that the batteries were maintained properly. The electric capacity stored in the batteries was normally sized to supply back-up power for 15 to 20 minutes under normal building load. That provides plenty of time to get backup generators running and ready to take over the load of the building.
An alternative to this approach is to use one or more dynamic flywheels to instantaneously supply electricity to the load. The monthly maintenance on flywheels is significantly less than operating conventional batteries. The flywheels may only cover the full load for roughly 15 seconds, but that is enough time to ride through most outages and to get the diesel generators, if needed, up to speed.
Since diesel generators are the cornerstones of data centers, it is important to keep them in impeccable condition. This not only includes providing scheduled maintenance, inspections, and periodic testing under load but also requires careful fuel treatment and fuel heaters. A properly maintained generator using warm diesel fuel can consistently start and be ready to receive load within seven seconds.
Early computer centers and data centers were maintained at 60°F. This not only made the general space relatively cold for employees but also required a huge amount of energy. A more novel approach is to provide direct cooling to the server racks and let the general space temperature reach 80 degrees or even higher. The reliability of the servers is unaffected, but the cost to cool the space is reduced.
Another novel approach to data-center reliability is to tie the various server loads within the building into a common electrical bus, called a “ring bus.” Within a data center building, there may be several independent electrical delivery points. Each of these delivery points is backed up by independent UPSs and generators. This creates the first level of redundancy. However, if the delivery points are connected to a common bus, then the bus can essentially serve as an extra form of redundancy. If the generator serving one delivery point fails, then the excess generation available on the bus can instantaneously fill in for the lost generator.
Data center designers, engineers, and operators maximize reliability. Many centers have implemented the features described above and have been able to maintain 100-percent reliability for many years. The adoption of these features will continue to grow, allowing for complete reliability at an affordable cost.
The goal is still reliability, reliability, reliability, but the data centers that can provide 100-percent reliability by using state-of-the-art designs in a cost-effective manner will be more competitive and provide greater returns for their investors.
John Caskey is on the board of directors of the Smart Grid Interoperability Panel and a senior member of IEEE.