Optimism about datacenter energy use
Datacenters aren’t the energy villains the popular press too-often portrays. While the “Real User Monitoring blog” has already punched several holes in this canvas, a few important points remain before we return next week to the details of keeping an eye on the performance your applications deliver to end-users.
At a highly-aggregated level, Jonathan Koomey of Stanford University documents that datacenter energy usage is falling below trend lines. This should surprise no one. Among other reasons, energy costs simply weren’t as important in earlier years of datacenter expansion. As the industry matures–and it has a long way to go–it’s only natural that “best practices” give a little more attention to energy efficiency. Energy efficiency still has enormous scope to improve, and it will.
Koomey astutely “… divide[s] the industry into four kinds of data centers …” Another component of industry-wide improvement is that the efficiency laggards, Koomey’s fourth category of “‘in-house data centers”, are declining in importance: “companies whose primary business is not computing” increasingly purchase SaaS (software-as-a-service), and thus shift datacenter operations to more expert and efficient hands.
At a much lower engineering level, much of the commentary about datacenter efficiency has made the mistake of proxying power consumption of whole facilities with what CPUs (central processing units) demand. This is misleading in at least a couple of distinct ways. First, many datacenters have been designed in such a way that cooling CPUs requires as much energy as the electricity they employ to move bits around.
Beyond that, there are plenty of operations where CPUs don’t dominate energy or dollar costs. Conventionally, CPUs co-operate with on-board memory, slower persistent memory or data storage, and networking, to accomplish computing goals. There are plenty of cases where data storage uses more energy than the computing which takes place within the CPU itself. As I began to describe yesterday, though, there are great opportunities to improve the efficiency of data storage, and particularly in its energy measures.
Another long-term trend favoring energy efficiency shows up in siting. As networking technology improves–this is another area that has improved rapidly, and also has plenty of improvement in sight–it becomes feasible to locate datacenters relatively far from their end users. Daniel Dern explains the potential location has to improve datacenter operations.
Energy matters. On a day-to-day basis, though, the best way you can improve the efficiency of your datacenter operations is through good professional devops practices, including application performance monitoring (APM) to ensure your users experience the application functionality they deserve.