No free cycles

MOST OF THE available CPU cycles in a company are wasted. That’s the logical basis of grid computing, but it promotes the fallacy that unused computing resources cost nothing. Just one midrange server can consume as much electricity as all the lights in your house combined, and it can pump out as much heat as a hair dryer.

Those unspent cycles raise your company’s costs for power, battery backup, and air conditioning. IT doesn’t pay the electric bill, so right-sizing computing assets isn’t the planning priority it should be.

Start by scavenging wasted watts from your servers. In a distributed setting, cut power usage by turning off underutilized systems. Systems management software can do this remotely, and it’s possible to bring machines up and down on a schedule or in response to usage fluctuations. If half of a two-node cluster is twiddling its thumbs waiting for the other server to fail, consider converting that hot standby unit to a cold (powered-down) spare. As infrequently as fatal system failures occur, reducing fail-over time from five minutes to 30 seconds may be too expensive. Cool down bigger iron by disabling or pulling extra CPU modules, shutting off redundant power supplies, removing unneeded memory, and yanking the power plug from unmounted hard drives.

The older a system is, the more power it wastes. I have a four-year-old multiprocessor Sun server that’s slower than a $700 desktop PC, but its power and cooling demands outpace its tiny cousin by a factor of 10. Every machine reaches a point at which its operating costs exceed the value of its contribution to the enterprise; such a system should be decommissioned, not repurposed.

The same goes for storage: A single 180GB Seagate Barracuda SCSI disk draws .53 amps at 12 volts. An array of 20 older 9GB Barracudas pulls 14.6 amps.

Still, there are ways to sidestep this drain. When you buy new equipment, fight the temptation to overconfigure. Instead of loading out every new server to limits of the budget, split the purchase. Two smaller servers will handle the load as well as one big one, and you can leave one server powered down until you need it. When comparing models, look for the sweet spot below the top of the line — if you can trade 15 percent performance for power savings of 40 percent, that’s a good deal.

There are other ways to reduce energy and cooling costs. Enable desktops’ suspend-to-RAM and hibernate modes. For less demanding roles, consider slower, cooler processors such as Intel’s Celeron, Advanced Micro Devices’ Duron, or the new super-low-power PC CPUs from VIA Technologies. Ban animated screen savers — they keep the CPU from throttling down to reduced power mode.

We turn off TV sets when we’re not watching them; doing the same with CRT monitors takes care of a major source of needless power drain and heat. At a time of painful cost-cutting, there’s no excuse for wasting money on unused and inefficient equipment.