Instead of Retiring Servers, Why Not Power Them Down?

Data center providers normally try to “right size” their facilities, either via consolidating older servers or making sure servers are running at peak efficiency through virtualization. TSO Logic has offered a third option: automatically powering them down when not in use.

TSO’s latest products in that regard are TSO Metrics, which monitors power use within a data center, and TSO Power Control, which actually manages that power usage.

Aaron Rallo, chief executive of TSO, said that he developed the solution after years of managing his own data center. The platforms are designed for data centers with noticeable peaks and valleys in terms of demand, such as holiday time when online sales spike: retailers and commerce sites are the company’s target market, along with sites which might see a surge of activity during the day as customers come online, only to experience an accompanying drop in activity at night.

“At Christmas time, when you’re really really busy, you may need something like 600 servers,” Rallo said in an interview. “In March, you don’t need all that.”

He added: “I found myself criticizing little expenses when I’d be writing $40,000 checks to the local utility each month. It became a little bit frustrating.”

But it’s not like power management has been ignored by the server industry. Dell has its Open PowerManage Power Center, HP offers Insight Control power management software, and NEC’s ESMPRO software does something similar. TSO’s solution works without a software agent, however, and provides a vendor-neutral solution.

TSO is announcing the general availability of the software; Rallo said the company is shooting for a 50 percent power savings versus un-managed servers. One customer, CGI house Arc Productions—which sees usage spike when it’s rendering effects onto video, and drops when it’s not—experienced a 56 percent savings, enough to earn the software a Green Enterprise IT Award from the Uptime Institute.

Turning a server on and off isn’t that difficult; the TSO software taps into wake-on-LAN and remote shutdown features to power up and down servers, using SNP and WMI. The secret sauce, Rallo said, is that TSO Logic’s software relies on application-level inspection to determine exactly how much of a data center’s power draw is going toward revenue-generating activities versus simply idling. This insight allows customers to save electricity without sacrificing performance, automatically controlling the power state of servers based on application demand. The TSO Metrics and Power Control software then offers the ability to manage the software using a “single sheet of glass” or device.

Rello also said that servers don’t necessarily need to be powered down completely to save power, as they often have different “sleep states” with varying levels of functionality.

TSO sells its software using either an opex (operating expenses model) or capex (capital expenses) model. Under the former, TSO will price the software by the number of virtual machines and virtual instances. The capex pricing scheme is less clear, although the idea is that customers will at least break even.

Next up? Most likely managing storage using the same techniques, with networking somewhere in the future, Rallo said.

Does TSO’s software make sense? In some ways, yes—saving power by remotely managing servers is a good thing. But datacenter admins may have some real concerns about lifecycle management, and the potential costs of spinning up and down disks. Of course, placing a server in some middle ground saves less power, but may mitigate those risks. So this might not be a solution for everyone.


Image: kubais/