The high cost of turning off your computers at night

The original article is the high cost of leaving your computers on over night.
They estimate that leaving computers on over night and during the weekends costs a company with a 1,000 computers $26,000 a year. They estimate that the US spends $2.8 billion dollars leaving computers on overnight and during the weekends.

While I am in favor of reducing carbon emissions there is another way of looking at it. Going from a completely off computer to having all the necessary apps and documents open takes a non-trivial amount of time, for the sake of argument lets say it takes on average 5 minutes. (I’ve seen environments where half an hour is a more realistic average.) Now lets also say that the ratio of employees to computers is 1:1, like in a typical office environment.
If you are paying minimum wage $6.55 per hour, and 1,000 employees are spending 5 minutes of each work day, 5 days a week, 50 weeks a year waiting to work the cost of lost productivity is 136,457.50 per year. This is a net loss of $110,457.50 versus the $26,000 saved on electricity.

If the 1,000 workers have an average salary of $70,000, then after the $26,000 electrical savings the loss of productivity cost is $675,041.66.

For a single employee who makes $1,000,000 a year the ~21 hours a year of lost productivity (5 minutes a day) has a cost of $10,375.

The article’s statement about in the US alone $2.8 billion a year is wasted on electricity for idle computers is very compelling. It becomes less compelling when you consider that the national average salary is $44,909 so if 108 million workers lose 5 minutes of every work day the lost productivity cost is ~$48.4 billion. This kind of over shadows the $2.8 billion in possible savings.


For full disclosure, let’s examine the biases’. The authors of this study sell software that auto-shuts down computers and restarts them in the morning, so this is a marketing campaign. However, their solution does reduce the lose of productivity costs, but not all computers can be or should be shutdown, and the complexity of implementing and maintaining a solution like this, not to mention the CYA issues the first time power management software nukes some important work justifies at least half, if not a whole IT professional. Using average IT salaries (with benefits and overhead) the cost of the head count adjustment is $38K to $75K a year. This is more than $26K, plus I checked and the software is not free, so the savings is more like $11K. You also need to consider the work loads on your servers and infrastructure since every employee will be switching on at about the same time and connecting to the email server and downloads emails, establishing connection to the application server, and databases.



Now most companies don’t have these automated solutions so the cost savings will be achieved using a memo ordering employees to shutdown, or having the cleaning staff shut machines down. This second option is insane, so it shouldn’t be tried. The first solution is also not very smart. I am being very generous saying it only 5 minutes a day in starting up time. Shutdown has a time cost too and often getting a computer ready to be shutdown can take far longer, than starting it back up in the morning. So for all you penny pinching CEOs, and CIOs think about both sides of the coin, since saving electricity could just push the cost off to salary. I agree that shutting down over weekends and holidays might not be a bad idea, but every night you really need to look at the ROI.

A better way to make money is to offer your company’s idle desktop machines for cloud and other distributed computing. The revenue is thin but this is real money instead of perceived savings, if nothing else donate the time to charity projects and write it off as a tax deduction. If you are worried about security or want to maximize opportunity potential, you can offer the CPU time to the company’s own R&D department. There are lots of brute force algorithms that provide better solutions than computationally tractable “elegant” work around/approximate algorithms.

Applications run the gamut from tool and die companies optimizing cutting paths and tool life, to retail stores running simulations of foot traffic and buying patterns in context of merchandise displays and store layouts, to city governments calculating traffic patterns as a way to plan infrastructure spending, using data collected from other cities to simulate future growth.