A client company is located in its headquarters location, but its Internet computers are colocated in one or more other buildings where they rent out space. For example, a website owner could place its website server on the premises of an Internet Service Provider (ISP), or an ISP could place its network router on the premises of the company offering switching services with other ISPs, or a company could locate in a carrier neutral facility with multiple ISPs available.
A COLO by Any Other Name ...Since this is such a new building type, the jargon or naming in the industry has not yet been standardized, and you might see this building described using any of the following terms: COLO, colocation facility, Internet data center (IDC), carrier hotel, server farm, webhosting facility, telecommunications hotel, Internet exchange, or cybercenter.
Some of these terms are more general and can include data centers that only serve one company in a building owned by that company. The COLO building is definitely a situation where one company owns the buildings and several different companies own the computers that occupy the rented space in the building.
Colocation customers receive Internet connectivity, power supply, cooling and humidity control, fire and security, and equipment such as cages, cabinets, and cabling. Generally, customers require 24/7 access to their computers and to the building. COLO buildings follow a general trend for companies to outsource key services that can be done more efficiently by others. COLOs also allow companies to come to market faster with an Internet product and with less use of their own capital and possibly to interact with other COLO customers in the same building.
These are durable, rugged buildings that must be able to "take a licking and keep on ticking." These buildings operate every hour of every day, with no planned shutdowns ever.
Predictions have been made that there will be half a billion Internet users by 2003. To that end, it's worth noting that while in 1997, the typical COLO building was under 25,000 sq ft, facilities under construction in 2001 are more on the 100,000-sq-ft scale. According to a San Francisco Chronicle article in June of this year, the industry had expected 400% growth in 2001 but will probably only grow at a 30% rate because of failures and slowdowns of dot-com companies.
The NIMBY's, Not In My Backyard folks, have found things to criticize in the COLO building industry. Taken as a group, COLOs need massive amounts of new power infrastructure, may depend on dirty diesel generators for backup power, and can supposedly dilute the lively restaurant and sidewalk scene in major metropolitan areas by building large buildings with no people in them. Yet, despite slowdowns on the stock market and community critics, the COLO has become a major new building type.
General Design RequirementsThe design process for a COLO requires a paranoid perfectionist, possessing both attention to detail and an active imagination to forecast disaster. What if the power fails? What if the natural gas service fails? What if the water fails? What if an air handler fails? What if an air handler fails at the same time that an electrical circuit goes out? How long does it take to restart the chillers after the emergency generators kick in?
Architecturally, the COLO building is intentionally low key. For security reasons, the building does not call attention to itself. A typical, plain rectangular design includes very few windows and a limited number of doors. Few staff are required to run the COLO building, so office space is minimal. Interiors have raised floors, large open areas, and high ceilings. COLO buildings are generally designed to continue operating during a moderately severe earthquake.
ReliabilityReliability is a major design concern of the COLO industry. Standards in the industry call for availabilities of six sigmas or five nines or six nines. A "six nine" reliability would be a facility that would be online 99.9999% of the time. By this standard, about 30 sec of downtime would be allowed each year.
Of course, reliability always comes with a price tag. The building designer must go to great lengths to avoid single points of failure. Backup systems, like diesel generators, may need to be power tested weekly. The conventional thinking on reliability is to go with redundant design. If you need three chillers and you install four chillers, you have N+1 redundancy allowing for the failure of one chiller. If you need three chillers and you install five chillers, you have N+2 redundancy. Some designers are now talking about N+1+1 redundancy. If you need three chillers, plan for one to fail or to be down for maintenance and for a major electrical circuit to go down. A COLO building should be able to go on operating with two major simultaneous failures like this.
Power ConcernsA 2000 study by the Allied Business Intelligence showed that of 150 gigawatts (GW) of new capacity required for all reasons in the United States, only 70 GW was being planned. Cost of energy is a concern as well as its quality and reliability. A study by Epoch Partners estimates 35% of all colocation charges are for utilities. A February 28, 2001 article in The Wall Street Journal cast some insight on the power problems caused by the huge demand COLOs can represent. According to the article, Pacific Gas & Electric denied power availability until the year 2002 for a 45,000-sq-ft Silicon Valley data center planned for construction in 2000. The article also cites an Exodus customer, Weather Underground, that pays $18,500 a month for its 60 servers in Santa Clara, CA. Exodus CEO Ellen Hancock projects revenues for 2001 of $1.35 billion. The San Francisco COLO, pictured on pages 72 and 74, has 85,000 sq ft of raised floor in a 230,000-sq-ft facility owned by AboveNet, a wholly owned subsidiary of Metromedia Fiber Network.
The growing trend is for COLO buildings to consider some form of on-site power such as a gas turbine or a natural gas engine. Some are choosing cogeneration to ensure power supplies. Third-party turnkey contractors offer financing as well as operating and maintenance staff and some even go so far as to offer 99.9999% availability.
The standard in the industry now is to rely on diesel generators for backup power generation. Efforts are being made for diesel generators to "clean up" their act so as to minimize the air pollution impacts of the COLO building, and different air districts employ a variety of requirements and/or alternatives.
Joe Fay, P.E., vice president of electrical engineering, Alfa Tech, Inc. (San Francisco), has designed hundreds of thousands of square feet of COLO space. He said that reliability and redundancy are the keys to good electrical design. According to Fay, COLO buildings need excellent grounding. To further increase reliability, most COLO buildings have power feeds from two independent utility substations. If there is a failure of power to a substation, or in the substation, or if the feeder from a single substation fails, full power will still be available to the COLO building from the other substation. However, adding the power feed from the second substation can represent a significant cost.
Power DensityPower densities in COLO buildings in 1997 were on the order of 30 W/sq ft. Power densities in 2000 were above 100 W/sq ft. COLO buildings in design are anticipating increased power densities in the future, in some cases as high as 300 W/sq ft.
For comparison purposes, a summer load in a California commercial office building might be 8 W/sq ft including associated air conditioning loads. The watts per square foot estimated for COLO buildings do not include air conditioning loads but just cover the actual connected electrical load for computer equipment and its peripherals. With these power densities, a 200,000-sq-ft COLO might require 50 MW of power. This single building would have an electrical load equivalent to that of 50,000 homes.
The driving force behind the rising power densities is new designs of computer servers from Sun and other computer companies. Like many kinds of electronic equipment, servers are shrinking in size but not necessarily shrinking in power requirements. For the most part, these computers are mounted one on top of each other in vertical racks. A common standard of measurement has come about of measuring the height of the racks in "U." Each U is 1.75 in. tall. Fairly recently, a server might have been 3 U tall. Today, that same server would be only 1 U tall. This allows three times as many electronics and their associated heat load to be mounted in the same rack.
MechanicalCOLO buildings have mission-critical air conditioning. The building cannot carry out its mission unless the air conditioning is taking out the heat from all the computers and associated electronic equipment. Depending on the power density, the computers will overheat and switch off within 5 to10 min after the failure of the air conditioning system.
Generally temperatures are maintained at 70 degrees F + 2 degrees with humidity levels maintained above 30% rh, to counteract the generation of static electricity, and below 70% rh. Filtration and air particle control are similar to a standard industrial setting or in good office buildings. Despite housing a lot of sensitive computer equipment, COLO buildings are definitely not cleanrooms. Because the buildings usually are well insulated, have very little glass, and contain immense amounts of heat producing equipment, often no heating is required. If the office and administrative area is big enough, it might need a small amount of heating to offset the envelope load. Unlike wafer fabrication areas, the COLO building is not particularly sensitive to vibration. Vibration standards would be similar to those of an office building.
The original designs for the small COLO buildings usually relied on a computer room air conditioner (CRAC). These CRACs would usually be DX systems with air-cooled condensers on the roof. While the CRAC takes up valuable storage space in the COLO (especially if it requires side maintenance access), it is flexible to design, modular, and easily offers some amount of redundancy. Oftentimes, owners will switch in subsequent facilities from CRACs with DX systems to water-cooled or chilled water-supplied CRACs. These units are quieter and more energy efficient. Several new designs are on the market to make CRAC humidifiers much more efficient in electrical use. These are designed to retrofit into older units to cut their peak electrical use.
More and more, the larger COLOs came to be designed with rooftop air handlers. The primary energy advantage, especially in places like the Silicon Valley, is that 100% outside air economizers could eliminate the need for chillers during a large part of the year. Some building owners have raised the question of cross-contamination of the facility by diesel exhaust during a power outage if the economizer dampers are open. A control sequence can be added to close the outside air dampers, as soon as the diesel generators are energized.
Even with rooftop air handlers, further energy-savings designs are possible. Switching to cooling towers and more efficient chillers have served to further reduce COLO operating costs for many owners.
However, the increased complexity of chiller/cooling tower systems must be addressed in order to achieve high degrees of reliability. Many COLO buildings install a water storage tank with several days' worth of cooling tower makeup water. This allows operation of the facility in case a water main breaks. In some cases, this cooling tower water storage tank is combined with a chilled water storage tank.
COLO buildings usually have a chilled water storage tank as part of their power outage strategy. When the power fails, the chillers will shut off. Depending on the manufacturer, there is a 5 to 10 min period for restarting the chiller, after the power cuts off. The chilled water storage tank is used to provide 10 min of full design chilled water capacity to serve the building while the chiller restarts. Usually the chilled water storage tank will have excess capacity so that it can serve the building both when the diesel generators come on and when they go off again.
The failure analysis in the mechanical systems of larger COLO buildings can be quite complex. Routing of piping and ductwork should always allow for multiple failures while still maintaining full building operation. What would happen if a pump failed? A pipe leaked? A duct collapsed? A power line was cut? As mentioned earlier, the risk analysis should allow for at least two simultaneous failures happening in different systems at the same time.
Controls and MonitoringEverything about a COLO facility is critical. Because of this, the amount of monitoring and speed of response is very intense. Jeff Ginn with controls contractor Energy Logics (San Jose, CA) discussed a recent COLO facility for which his firm was the contractor. The building owner was Digital Island, a COLO company owned by Cable & Wireless. The facility was 80,000 sq ft located in San Jose, CA. The building was initially occupied in 2000. The building management system manufacturer was Andover Controls.
When the facility is fully occupied with customer computer systems, the build out of the building management system (BMS) is expected to be 8,000 to10,000 points. For comparison purposes, an Andover installation at Boston University handles 75 buildings with 3 million sq ft with only 12,000 points.
The points on the system are expected to be distributed as follows:
- Fire alarm, preaction sprinklers, and FM-200 system: 1,500 points;
- CRACs and power distribution: 5,000 points;
- Generator and fuel systems: 750 points;
- Chillers by The Trane Company: 500 points;
- Rooftop package units by Carrier: 500 points;
- Variable-air volume (vav) boxes: 250 points; and
- Security: 200 points.
Security measures include fingerprint scanners, biometric hand readers, card key access, and video surveillance with digital storage. The building also supports systems for intercom, paging, fire alarms, elevator controls, and personal duress transmitters.
Digital Island placed a big emphasis on speed, reliability, and integration. Alarm points must be monitored quickly with short response times so that corrective action may be taken immediately. The alarms are passed through the network operations center (NOC) so that it is available immediately to Digital Island company offices around the world. The system's e-mails alarms to company employees' computers, cell phones, and pagers.
Since preventing and solving problems quickly is so important for this type of building, the monitoring and control system itself must be extremely reliable.
Integration plays its part in both increasing the speed and reliability of the system. In this design, the various subsystems do not report to proprietary computers but instead each subsystem connects to a dedicated Andover controller. The fire alarm, chiller, air handler, vav box, generator, leak detection, power reliability, and security systems are all integrated under Andover controllers and a single graphic user interface. Modbus is used as the communications protocol.
By having the integration occur at the controller level instead of the computer level, there is increased speed and reliability in the system. Also, hardware protocols are easier to work with and more standardized by connecting to the subsystems lower down in the hierarchy.
The facilities staff only has to learn and maintain one software package to monitor and control fire, security, mechanical, and electrical systems. A password-enabled user, located anywhere in the world, is able to access temperature, humidity, fire, security, and power data online. A similar system in a high-rise building enables each office occupant to go onto the Web and obtain his/her office or zone temperature reading. This has reduced the number of hot and cold calls to the building maintenance staff.
Plumbing, Fire Detection and SuppressionCOLO buildings require a certain amount of restroom space, roof drains, break rooms, water coolers, and the like. Great efforts should be expended during design to try to isolate the rack areas with computers from coming into contact with any possible water leak. Usually multiple water leak detectors are installed under the raised floor and anywhere else they might serve as an early warning system against the possibility of a water leak. In addition to standard plumbing, the COLO has an intense amount of air conditioning. This leads to a great deal of water piping for humidification and also for draining off a large amount of condensate from air conditioning units.
COLO buildings will usually need to have a great amount of diesel fuel stored on site to feed the emergency generators. This may mean that there will need to be oil tank farms serving the massive banks of generators that keep the COLO building running. Instead of several hours of backup fuel, there may be several days' fuel supply so that the COLO can be kept running come what may. Operationally, there will need to be a fuel restore contract so that the COLO gets first crack at a refill of its diesel storage tanks after a rolling blackout or a major regional power outage.
Also of great complexity is the fire protection design. Many COLOs have both preaction fire sprinklers as well as a gaseous extinguishing system like a FM-200 system. Fire and smoke detectors are specified at their most sensitive so that any generation of smoke can be detected before the smoke damages the sensitive computer equipment.