Just when data centre managers are wondering whether to replace their equipment with leaner machines, or sweat the assets a bit longer, Peter Judge hears of another idea. Cooking your servers
If you want your data centre to make less impact on the environment, should you carefully preserve your equipment to make it last as long as possible – or should you burn it up more quickly than that?
The answer sounds obvious. The government is telling companies to “sweat” their assets, and keep equipment for longer – and 26 percent of them seem to be following the advice. But is that really the best thing to do?
Dell, Intel and others have long said that new hardware is so power-efficient, the savings on the electricity bill alone justify junking servers more than two years old. I’ve always been suspicious that this argument lines up so perfectly with what Intel and Dell want – just as I’m dubious that they really think their servers will save the planet overall – even if their new servers are very efficient, I don’t think they plan to actually sell any fewer of them.
But here’s another angle – a Sun executive, in a very interesting webinar is arguing that it’s a good idea to turn up the temperature in your server rooms – even if that actually makes a measurable difference to your equipment’s lifetime… and actually burns it out sooner.
There is now an Energy Star label for servers, which ought to make it easier to plan to use less energy in the server room, but some people have pointed out weaknesses in the label, and other agencies have competing certificates.
But the Energy Star comparison includes “idle” power in the Energy Star, which isn’t necessarily useful, says Phil Morris, chief technologist for Sun’s data centre efficiency practice, in a presentation entitled Power Management and Data Center Efficiency. Since servers are most efficient when they are used at full stretch, and most data centre managers try to use them that way, “it’s like compare a huge SUV with an energy-efficient car – while they are sitting at a stop light.”
But. although you can’t expect to turn off servers for long periods, like desktops, he thinks it might be a good idea to turn them off when possible… even though this may actually shorten their working life.
Servers are designed for continuous use, and might have a mean time before failure (MTBF) of twelve years if left on continuously. This could reduce to nine years, if they are turned off every night.
However, this is still substantially longer than the five year replacement cycle that Morris represents – and energy savings of $1000 per server over that time are worth having.
The details are all in his webinar, but here’s the point. Hardware makers used to deliver the highest performance hardware. Now, high efficiency is also prized – but data centre managers should get the choice to run their hardware at any point between these two extremes, depending on the needs of their customers, their chief financial officer… and the all important CRC budget.
There really does seem to be a new way of thinking about data centres evolving. And data centre managers should have the choice of sweating their assets… or cooking them.