Hardware Today: Keeping the Server Room Juiced

Monday Aug 29th 2005 by Drew Robb
Share:

No matter how state-of-the-art your equipment is, without power it's essentially an expensive paperweight. Thus, a UPS is a critical component of any server room. The latest units have scalability and servicing models similar to those of servers

Servers have gotten continually smaller, more modular, and more reliable over the years. The Savvio family of disk drives from Seagate Technology, for example, cuts the platter size to 2.5," slashing power consumption by 40 percent and allowing more devices to occupy the same rack space. At the same time, they maintain a 1.4 million hour mean time between failure — that's nearly 160 years.

No matter how reliable the server, however, the disk stops spinning the second power is cut, and a rack-dense design means a great deal more power must be run to that location. A single rack of IBM p5-575 servers, for example, consumes more than 40 kilowatts, and HP crams 192 of its 35P dual processor blade servers into a standard 42U rack.

To ensure servers have a reliable supply of clean power, infrastructure vendors have been introducing more compact, modular, redundant, and hot-swappable power units. Concerns over rising energy costs have contributed to changes in power equipment, and vendors have taken steps so servers better adapt to the customers' changing needs. Rather than building in a lot of overhead from the beginning, enterprises can initially install only what they need, and add or move things around later. This cuts down the initial capital expense and provides greater flexibility.

"While many servers have traditionally been scalable, today's servers require you to do more planning when deciding on your power protection strategy." — Steven Carlini, Director of Product Management, American Power Conversion.

"While many servers have traditionally been scalable, today's servers require you to do more planning when deciding on your power protection strategy," says Steven Carlini, American Power Conversion's (APC's) director of product management. "Companies can grow in an unorganized fashion and purchase a uniform UPS [uninterpretable power supply] for each rack of IT equipment. But that is not the best way to optimize TCO, business continuity, agility, and [it] complicates life cycle management."

In essence, APC has adopted a strategy similar to the one OEMs use with clusters and blade servers — build out only as much as is needed at the time, and add more units as requirements ramp up. According to Carlini, APC has a full portfolio of Symmetra N+1 UPS available from 2kVA up to 1.6 Megawatts. He emphasizes that enterprises can "right-size" the UPS for their specific application and plug in more modules, as required, for increased power, runtime, or redundancy.

In addition to cutting capital costs, this modular design cuts down on power losses, since UPSes are more efficient when operating near capacity. Carlini also notes that this approach reduces cooling costs because lost power is converted to heat.

Hot Swappable

In addition to becoming more modular, power units are paralleling server designs in their serviceability. Just as one can swap out blade servers without powering down the rack, APC is manufacturing hot-swappable power supplies and UPSes to eliminate no-service interruptions when working on the infrastructure. The same ease applies to power cabling.

Universal Electric Corporation (UEC) of Bridgeville, Penn., for example, has a product line called Starline Track Busway, which does away with the need to run dedicated cables from the breaker box to each rack unit or server. Instead, an organization installs the Starline Track Busways above each row of cabinets. To bring power to the servers requires inserting a plug-in box with a drop-down power cord and one or more receptacles, and giving it a 90-degree turn. To remove power from one location and move it to another necessitates the same — a simple twist-and-pull operation.

"Code requirements say you have to remove an under-floor power cable whenever the load or computer is permanently removed, which can be quite a job" says Joel Ross, UEC's president. "But with our product you just twist the electrical plug-in box 90 degrees, and it drops right out. Then you can relocate it, plug it back in, and there is no cabling to reroute."

Since the circuit breakers are built right into the plug-in units, there is no need to shut off power at the breaker box. He adds that these busways reduce the amount of cables a company must run and the overall number of power distribution units that must be deployed. The units scale from 40 amps up to 225 amps/600volts. Ross says that although the initial hardware costs for the buses may amount to slightly more than just running a cable, they can lower labor costs, and in some cases they may eliminate the need for raised floors.

"When you want to relocate or add additional units in the future, you can just plug a box into the busway in a matter of seconds," says Ross. "Otherwise, you have to run a long cable back to the circuit panels, which is costly in terms of material and, especially, labor."

>> Power Maximization

Getting in Condition

Getting the power to the servers is one problem. Another is making sure it is noise-free and at the right voltage. While UPSes and surge suppressors take care of the large spikes that can damage the equipment, one must also address the small voltage fluctuations. According to a study by Bell Labs, blackouts and surges higher than 200 volts caused only 3.8 percent of power-related issues, while voltage sags and voltage surges less than 200 volts accounted for 96.2 percent of the problems.

But power is not a stand-alone characteristic — more power means more heat. Concentrating power usage concentrates the heat and raises problems getting the airflow where it is needed. The resulting extra cooling costs may then exceed any savings from using less space.
"Any spike greater than 1 volt confuses the logic — the microprocessor reading it as a 1 rather than a 0 — resulting in screen lock-ups, time-outs or delays," says Bahram Mechanic, CEO of Houston-based SmartPower Systems. "Power problems caused by small surges, spikes, and sags in the electricity supply cause 15 times more problems today than viruses."

Surge protectors do a good job against lightning strikes and UPSes prevent blackout losses, but neither is designed to guard against these smaller, more common fluctuations. Isolation transformers filter out the smaller variations, but they are large and expensive. SmartPower has a newer alternative called transformer-based filtering (TBF), which uses transistors, thyristors, capacitors, and relays, together with a small transformer, to condition the power. The devices weigh 17 ounces and cost about $100, making them affordable for even the smallest server closets. Although SmartPower sells UPSes, surge protectors, and isolation transformers, according to Mechanic, "servers, workstations and networking gear can best be protected by using transformer-based filters."

Avoiding Meltdown

New power products make it easier to support the growing popularity of rack-dense servers. But power is not a stand-alone characteristic — more power means more heat. Concentrating power usage concentrates the heat and raises problems getting the airflow where it is needed. The resulting extra cooling costs may then exceed any savings from using less space. So being overzealous is counterproductive.

"We recommend that clients keep their power standards in area of 50 W/ft to 100 W/ft and strive to have an average of 4 kW per rack, including 60 percent for cooling," says Gartner Research Vice President Mike Bell. "When they get over 100 W per foot they start getting into a dysfunctional situation where the heat output is so intense that they see their power costs go vertical."

Share:
Home
Mobile Site | Full Site
Copyright 2017 © QuinStreet Inc. All Rights Reserved