Companies are becoming increasingly aware of the high costs of maintaining and cooling servers: A mid-tier $2,500 server can cost as much as $2,020 to power and cool annually, according to the Uptime Institute. Thus, demand for ways to curb those expenses is rising as well.
Strategies such as server consolidation and virtualization are gaining momentum as a means of reducing server head count. Once you're down to as few servers as you need to meet peak demands, however, there are still inevitably times when you're wasting substantial energy powering idle machines that are waiting for something to do.
That's the case being made by Cassatt, a relatively young company based in the US, as it pushes its Active Response offering. The software tool is designed to help admins make more efficient use of their datacenter resources -- and it's a big step toward on-demand or utility computing, in fact.
With the new Version 5.1 release of Active Response, Cassatt has added a very enticing capability: Admins can now set up policies such that application resources -- physical or virtual servers, software, and network resources -- will automatically adjust to meet real-time demand or service levels. The idea, in a nutshell, is to have just enough servers running simultaneously to satisfy users' needs, thus saving money on keeping excess servers from needlessly drawing expensive energy and cooling.
To understand the possibilities here, let's consider a scenario where your company has, say, 500 Web servers dedicated to selling widgets. A couple times a month when widgets are on sale, demand shoots up and the servers reach 85 per cent utilization. Most of the time, it's more like 50 per cent. After hours and on weekends, though, it drops down to 15 per cent. (These numbers are purely hypothetical, given the difficult to track widget demand in today's tough economy).
The problem here is, even when the servers are at 50 per cent or 15 per cent utilization, they're drawing energy all the time. Research has found that servers burn more than half of their rated power while idling - which in turn amps up the need for additional cooling, burning further power. Now you might view keeping all those machines powered on all the time as insurance that you can meet service demand if it suddenly shoots up unexpectedly. But given the rising costs of energy, it's an expensive premium you're paying.
With Version 5.1 of Cassatt's Active Response, an admin could create policies to power on physical servers and create virtual ones one the fly, only as demand dictates -- and to power them down when demand drops. For example, you might set policies such that when your first 20 servers reach 60 per cent utilization, another 10 will turn on. When those 30 hit 60, another 10 will turn on, and so on. When utilization drops to whatever level you choose, you set the machines to power down accordingly.
Savings here can be substantial: If you've got 500 200W servers, your energy rate is ten cents per kWh (the US average), and you can power down machines just 25 per cent of the time, you can save nearly $50,000 per year on power and cooling expenses. Those savings scale depending on how many servers you have -- as well as how high energy costs soar.