Architecture and Hardware

Software-Defined Power Cuts Costs

A representation of software-defined power.
Software-defined power is about creating a layer of abstraction that makes it easier to continuously match datacenter resources with changing needs.

Datacenters, especially hyper-scale cloud providers, have demonstrated the higher efficiency, higher rack density, and lower capital outlay that can be achieved through virtualization of software-defined servers (SDS), software-defined storage, and software-defined networks.

According to IDC research director Jennifer Cooke, software-defined power (SDP) is next to be virtualized in the modern software-defined datacenter. SDP creates a layer of abstraction that makes it relatively easy to continuously match power resources with changing power needs.

The technology was created a decade ago by Power Assure, a Santa Clara, CA-based datacenter infrastructure management startup, but was ahead of its time, since the traditional datacenter model to this day intentionally over-provisions power to every resource in order to facilitate minimum down times and faster execution speeds. However, the next generation of so-called "hyper-scale" datacenters—the cloud-based successors to on-premise datacenters—are turning to SDP to meet their dynamically changing power demands.

Hyper-scale datacenters lower costs to enterprises by elastically scaling their architectures as demand fluctuates. As a consequence, enterprises pay only for the computer time they use, and at a lower rate than an internal datacenter. By seamlessly provisioning server, storage, and networking resources to its many users' needs, hyper-scale datacenters realize a computing environment that is robust and scalable for both local and distributed program execution. Hyper-scale datacenters are also virtually essential for today's power-hogging workloads such as blockchain data mining, machine learning, anything using graphics processing units (GPUs), and Internet of Things (IoT) devices.

To meet these goals while making a profit, however, is requiring hyper-scale datacenters to rethink the provisioning and realtime distribution of power. According to the Ponemon Institute, an organization that "conducts independent research on data protection and emerging information technologies," power and cooling account for 40% or more of a traditional datacenter's annual costs. The U.S. Department of Energy (DoE) predicts datacenters nationwide are facing a six-fold increase in power usage by 2020, anticipating their electrical consumption will grow from 30 billion kilowatt hours (kWh) in 2000 to as much as 200 billion kWh (or as little as 70 billion kWh, for power-thrifty hyper-scale datacenters).

"Software-defined power automates energy efficiency in the datacenter, using AI [artificial intelligence] and machine learning to predict and react to power usage profiles, rather than relying on human-reactions to provision power manually," explains Cooke. "Datacenters have always been interested in power efficiency, but until the hyper-scale datacenter, they have not been spending enough on virtualizing software-defined power, especially at the edge where energy is dear."

According the DoE, without power efficiency steps such as SDP, as much as 20% of grid power will be consumed by datacenters in 2020, compared with 2-3% today. Amazon, Google, Microsoft and other mega-users of electricity have tried to solve the problem by renting idle datacenter resources to cloud users, or offering batch-job services like Mechanical Turk, which run whenever resources would otherwise be idle.

Such measures do not minimize energy consumption as their main objective like SDP, whose core idea is to stop over-provisioning each server, storage unit, and network. Instead of oversized power supplies with enough capacity to handle peak consumption periods, which are underutilized the majority of each day, SDP provisions power dynamically in real time.

"Today's mass over-provisioning of power leaves up to 60% of paid-for datacenter electricity underutilized," said Virtual Power Systems (VPS) chief executive officer (CEO) Steve Houck, a former vice president and general manager at VMware. "Software-defined power or SDP solves that problem with the same kind of virtualization that VMware has already used to revolutionize the efficient utilization of server-, storage- and network-resources at datacenters."

As mentioned, Power Assure was the first company to bring SDP to market, back in 2007; the company raised over $60 million to popularize the technology, only to go bankrupt in 2014. Companies like VPS say Power Assure was not only ahead of its time—since power management was not yet a necessity—but it had not perfected intelligent control of energy (ICE), a term that denotes the capabilities required to realize SDP with AI and predictive machine learning. SDP requires ICE-compliant hardware instrumentation inside servers, storage units, and networks to enable power consumption to be closely monitored and fine-tuned by software in real time.

"Traditional datacenters are way over-provisioned and really need comprehensive software that monitors and manages the power usage of all their equipment so they become more efficient without the risk of slower execution speeds or increased downtime," said IDC's Cooke. The energy consumption problems facing the hyper-scale datacenter economy of today proves that the time for SDP has finally arrived, she adds.

"Datacenters only react out of necessity, which was the case a decade ago when virtualization made usage of servers, storage, and networks efficient enough to be profitable," said VPS's Houck. "Today, the biggest hyper-scale datacenters can't even pull enough megawatts off the grid to power their biggest installations without SDP."

There are other software-defined power specialists besides VPS, but they cannot use the phrase "software-defined power" because VPS acquired the trademark for that phrase, previously held by Power Assure.  Competitors include IBM's "green infrastructure" instrumentation and software manager, which the company uses for its own servers, storage units, and networks; QHi Group's Exertherm thermal monitoring hardware, combined with its Vigilance Energy Management software, as well as Egenera's Xterity, Hitachi's Vantara, and Siemen's Clarity datacenter management software packages.

VPS offers two main services. The first is to make an existing datacenter ICE-compliant by installing its SDP software along with a dashboard that uses AI and machine learning to automate real-time power efficiency algorithms, or alternatively integrates the same SDP algorithms into the datacenter's existing management dashboards. Its second service is for new datacenters, into which the company installs SDP-compliant servers, storage units, and networks; VPS's SDP dashboard then runs on a single x86-based I-Server to manage the AI and machine learning algorithms that virtualize power.

The company also offers massive integrated lithium-ion battery banks (which it calls ICE blocks) that provide standby power for peak-shaving load management of datacenter power.

R. Colin Johnson is a Kyoto Prize Fellow who ​​has worked as a technology journalist ​for two decades.

Join the Discussion (0)

Become a Member or Sign In to Post a Comment

The Latest from CACM

Shape the Future of Computing

ACM encourages its members to take a direct hand in shaping the future of the association. There are more ways than ever to get involved.

Get Involved

Communications of the ACM (CACM) is now a fully Open Access publication.

By opening CACM to the world, we hope to increase engagement among the broader computer science community and encourage non-members to discover the rich resources ACM has to offer.

Learn More