Consider the Power Supply
Nikola Tesla discovered alternating current electricity in 1883, allowing electric power to be transmitted over long distances. Fast forward to 2001, and data centers everywhere still run on AC. However, not all electrical systems are created equal. How your wiring is installed and managed has a lot to do with your systems' reliability.
Although construction policies vary, many companies do "lowest bidder" contracting for an entire building, which unfortunately includes the data center. Your architect will specify the appropriate power systems for the data center, but will they be installed correctly? Loose ground wires, polarity mix-ups and undersize conductors can cause major problems, and not only in the data center itself. Power problems can spread, affecting your end users' workstations. If you can influence the decision to have data-center contracting done by specialists, you'll be ahead of the game.
For those of you building a new data center in an existing building, be sure to consider whether the building's main power supply is adequate. If possible, bring in a public supply solely for the data center. That will help isolate the data center from power problems in the rest of the building. Early in the process, check with your local power utility to see what kind of lead time it needs for installation. Many local utilities sell generators and UPS equipment, so you might also ask about extra services to help ensure power quality. Finally, remember that deregulation can affect your quality of service, as California has so visibly demonstrated, so ask what impact local deregulation could have on you.
Experience has taught us that no matter how good the quality of a given workstation or PC, the unit's operation will be problematic if the AC supply is not within tolerance. This is one of the trickiest problems to diagnose and cure, given that upper management tends to treat reports of power problems with skepticism. Even technical managers tend to look askance at you when you explain that building power is the cause of some problems. One response you're likely to hear -- "Well, I don't see the lights flickering" -- merely highlights the fact that many network administrators don't fully understand the intricacies of AC power. Selling the idea of power problems that you can neither see nor detect without expensive test equipment is difficult.
Separate circuits and individually grounded computer-grade power sources are necessary. The smallest details in a data center's grounding scheme can make a big difference. Your ground potentials need to be tested. Contrary to popular belief, grounding to a building's water supply is not a smart move, and in many municipalities it's not up to code either. Similarly, if too small a grounding wire is used, there may not be enough ground. Simple things like failing to scrape the paint away before locating a ground on a building structural component can impede grounding efficiency.
Use A Pro
If you don't know about AC power systems, hire someone who does. Find out if your company, or the collocation center you're considering, has ever had a power survey done. If so, find out if surveys are taken periodically. You might be unpleasantly surprised by the answer. Power problems can trigger a slew of avoidable malfunctions in both your servers and your network equipment, and putting your fate in the hands of your local electrician is a crapshoot.
Even though it's a good idea, most companies never hire a power-quality professional. Remember, this person is not an electrician, but rather an electrical engineer who understands data-grade power and knows how to make your systems operate at that level. Quite a few architectural firms and most local power companies offer this service. An engineer will use professional-grade, power-quality monitors, such as those offered by Fluke Corp. and Dranetz-BMI, that record voltage transients, harmonic distortion and waveform degradation. Monitors will be hooked up in various parts of your building for hours or days, recording data for analysis. Armed with this information, your power-quality professional can pinpoint problems and recommend solutions for your local electrician to implement.
Watch Out for Transients
Transients, or surges, are sudden pulses of current or voltage in an electrical circuit that last a very short time, usually milliseconds. While popular opinion blames electrical storms as the main source of voltage transients, the truth is that transients are caused by a variety of factors, including switching by a company's utility grid; load-switching by other companies in the area; and use of variable-speed motors, such as those found in HVAC equipment, photocopiers and laser printers.
Another distressingly common source of transients is fluorescent lighting. That's right, the simple fluorescent light on your desk -- plugged into the same outlet as your computer -- can cause big problems. In fact, when a 4-foot fluorescent light is switched off, discharge of the fluorescent tube can cause a voltage transient in excess of 1,200 volts. Other likely sources of transients are space heaters and refrigeration equipment.
It's hard to get quality transient protection. It doesn't help that your local retail discounter sells in the lawn and garden department devices labeled "surge-suppression strips" that would be more aptly named "short protectors." These prevent fires, not damage to your computer equipment. Make sure that you don't have this kind of substandard device in your data center.
A couple of technology-specific tools are commonly used in true TVSS (transient voltage surge suppression) products. The first, and probably most common, is an MOV (Metal Oxide Varistor). When an MOV receives a transient, it clamps down to stop the transient. Clamping voltage is the level at which a TVSS device will begin to shunt a voltage transient to ground to protect the equipment attached to it.
One of the known issues with MOVs is short failure. Sometimes when a large transient hits an MOV, the device will fuse shut and not open as it's supposed to. When this occurs, the transient is passed, and the resulting short can be a fire hazard. Another subtle and quiet problem with MOVs is degradation. MOVs are rated for a certain number of transients at a specific voltage. After this number has been reached, the clamping voltage will rise above acceptable levels. In other words, your surge suppressors wear out.
There is no way to test the remaining life in an MOV surge suppressor, and this means that lots of old MOV devices in use are no better for surge suppression than the aforementioned power strips from the lawn and garden department. Treat your MOV TVSS equipment as you would an inexpensive smoke alarm with permanent batteries: Throw it out every year and buy a new one. If you have been paying only $30 for your TVSS devices, you're buying cheap MOV technology units that degrade quickly. Remember, there is no way, with the exception of a destructive test, to tell how far an MOV device has degraded. Gimmicks like "Protection Good" lights are just that: gimmicks, used by manufacturers to sell TVSS devices.
Although you can put a large number of MOVs in a TVSS device to make its maximum protection life five years or longer, this setup does not solve the problem of MOVs' fusing to a short condition. When you purchase a TVSS device, the product literature will usually indicate the type of protection used. If it does not, call the manufacturer and find out.
Generally, TVSS devices also tout "guaranteed protection," which means that if the device fails to protect your system from a transient, the device manufacturer will replace the protected system. Take this guarantee with a grain of salt. If you don't register your surge protector and send in the model, make and serial number of the item to be protected, there's no guarantee. You are also required to submit the broken item so the manufacturer can determine whether it suffered damage from a transient. The whole process is so obfuscated that you're unlikely ever to see any return.
Another surge suppressor, which does not have the degenerative characteristics of MOVs, is the SAD (silicon avalanche diode). An SAD device will not degrade with use, and it also has a lower clamping voltage than MOVs on a 120-volt line. SAD devices will clamp at 230 volts peak, while MOVs typically clamp at 330 volts peak.
For an additional layer of defense, you may want to install lightning protectors. These large panels with massive grounds generally are placed between the public power supply and the main power panel in a building, and shunt most of the lightning away from the main power panel. Although lightning protectors are destroyed during a direct hit, that's much better than the alternative. Your good-quality TVSS device should be able to handle any remaining voltage transient resulting from the lightning strike.
If your data center or collocation facility is in an area where electrical storms are common, you may want to put a TVSS device on your communications lines. Data TVSS devices are expensive, but Ethernet, serial and telephone lines can receive voltage transients from nearby lightning strikes. This phenomenon is caused by electromagnetic induction. We have seen communications chips with their tops blown off while the power supply and the rest of the machine remained in good working condition.
Whatever protection you use, ensure that you have high-quality devices that will not degrade in your office environment. Don't curse a TVSS device that appears to be dead. If it is, it probably saved your equipment. The old saying "You get what you pay for" is especially true when it comes to TVSS devices.