Silence your servers: It's quiet, it's green, it's the rack o' my dreams
- 15 May, 2008 10:57
When I refer to my lab, I use the term loosely. It's a 10-by-10-foot working space whose smooth walls channel the sound from every device with a fan straight into my ears. I share that room with every server I use and test. Of these, an 8-core Xserve is the only box that stays on 24/7, and I wish I could say I've gotten used to the noise. I haven't. While the Xserve idles at a pleasant noise level, as soon as any computing load kicks in, the fans spin up. When they do, they find a frequency resonant with the part of my brain that tells me that if I value what's left of my hearing, it's time to leave the room. The necessity of working with rack servers that get louder with each generation has made noise the primary governor of my workflow.
Of rack servers, Xserve is relatively quiet. Apple's design favors ergonomics, but this Xserve is configured with 8GB of RAM. For contrast, consider the four-socket, 16-core, 32GB 1U Barcelona rack server that AMD recently shipped to me. At idle, that machine is as loud as Xserve is at full tilt. My 16-core Xeon rack server is no better. I honestly can't live with them. I was ready to stick them in my garage, sucking wind from a portable air conditioner. The combined racket would be intolerable.
There are three things that I set out to save: My ears and my power draw. To rescue my hearing, I shopped endlessly for noise reduction solutions, from sound-absorbing pads that stick to the wall to refrigerated racks that are, more or less, refrigerators. Sound-absorbing this and noise-scattering that, when they're pitched as solutions meant to work outside the rack enclosure, are glorified packing foam. The cost of cooled enterprise racks is so outrageously high that an employer would have to judge the expense greater than the value of one's hearing. Noise is not taken seriously as a workplace hazard for white-collar workers in the United States. Mark my words; it will be. But even those enclosures that seal for self-cooling are built for noncooled and outdoor environments, neither of which is my problem.
That long search brought me around to a company I've known about for a long time, but didn't associate with solutions suitable for enterprise use. After a long and edifying discussion, GizMac, a company that really needs to work on its name, agreed to send me an XRackPro2 sealed rack enclosure. GizMac was careful to set my expectations. XRackPro2 is not, the company warned, a noise-isolating cabinet. It reduces noise, I've learned, with varying effectiveness depending on the type and amount of fan noise generated inside the rack. But I'll tell you this: I packed an 8-core Xserve and two 16-core machines in a 6U XRackPro2. When I powered them all up, the noise was so overwhelming as to make a telephone call impossible from anywhere in the room. Until, that is, I shut XRackPro2's foam-sealed front and back doors. I sat there opening and closing the doors for quite a while, marveling at the difference in noise levels. I also discovered that the forced airflow through XRackPro2, with a filtered intake underneath the enclosure, where the cool air is, and a pair of huge AC fans mounted to the rear door, made the server fans spin considerably slower, further helping to control noise. GizMac chose the fans for the rear of XRackPro2's cabinet well. They are barely noticeable.
XRackPro2 makes a jaw-dropping difference in rack servers' noise level, but by itself it isn't enough. The 6U XRackPro2 renders an 8-core Xserve silent. Even in the XRackPro2, the noise from three servers churning under high workloads falls from painful to safe, but in a 10-by-10 space, noise-reducing headphones are still occasional companions.
There is, however, an unwelcome contributor to server noise -- specifically, the noise that servers generate when they're shut down. A shut-down server's power supply fans keep spinning to supply power to subsystems that are always operating, like LAN interfaces and system management controllers. Why the fans have to spin so fast to supply so little power is a mystery to me. In any case, I consider the maximum acceptable power draw for an unused server to be zero watts, and the maximum noise level to be complete silence. This is achieved by yanking the AC plug from the wall, something that few servers can do for themselves. But there is a way to do it.
Join the Computerworld Australia group on Linkedin. The group is open to IT Directors, IT Managers, Infrastructure Managers, Network Managers, Security Managers, Communications Managers.
Optus goes over the top with VoIP service
Turnbull asks how the NBN got that way
U.S. retailers insist on PIN requirement in smartcard rules
Yelp speeds database access with flash storage
Thanks a million, Drupal