When speaking at public forums I often relate the operational aspects of running a data center to those of running one’s home. Most people can easily relate to the home environment in large part because they have a financial stake in the decisions they make. Unfortunately this is often not the case in a data center.
Take how we manage the lights around our home and compare that with servers in a data center. Many homes have exterior lights, either for security, convenience, aesthetics or a combination thereof. These lights can be equipped with; motion sensors, timers, photo-detectors (day/night), or a simple on/off switch. For the most part the lights are critical at specific times, so we ensure they are on when needed. Not to be confused with the annoying neighbor who leaves their lights on 24x7!
Yet when we explore the inner workings of a data center, where frequently the interior lights are indeed controlled with motion sensors (that’s a good thing), the servers are on 24x7 as if teenagers ran amuck when no one was looking. Not exactly what you would expect from an industry that has come under public scrutiny for the sheer amount of energy it uses on an annual basis.
And why is this you may ask? I suspect that Fear, Uncertainty, and Doubt, the all important “FUD factor”, is at work as IT and facility managers, by the very nature of their jobs, are risk adverse. In this case their overarching concern is, that much like a light-bulb, the server will fail to function when asked to turn-on. So rather than risk the possibility that a small percentage of servers may fail on turn-on, data center operators prefer to leave them running 24x7 throwing money, energy, and human resources out the window in order to support this wasteful practice. A practice, full of data center “myths”, that is no longer justifiable with current technology and business systems.
Myth number one – too many severs will fail and we won’t be able to support our SLAs! (Where are the Myth Busters when I need them?) Answer, part one, failure rates tend to be bathtub shaped, meaning a high degree of infant mortality, generally found in test/dev labs, start-up, and first few weeks of operation. After that failure rates tend to stay low for an extended time, then start to rocket up at end of life. Part two, your suppliers will provide a guaranteed SLA for replacement units on-site, often in less than four hours and perhaps two hours, in major metro markets. Even with OS and apps pre-installed!
Myth number two – servers were not designed for repetitive cycling. Answer, part one, contact your server OEM and get the real data! In a five-year business cycle you only need to turn servers on/off 1,400 times. Part two, Fujitsu has now introduced a server designed specifically for repetitive on/off cycles. I bet they won’t be the last!
So who is going to be the first data center operator to start turning their servers off at night and over the weekends? My guess is this will be a common practice in five-years or less.
Email me with your thoughts.