Toward an Energy-Efficient, High-Performance Datacenter
With news that atmospheric CO2 concentrations are due to reach the uber-critical 450 ppm mark in just a couple decades, the need for concerted mitigating action is more urgent than ever. Despite partisanship, consensus has coalesced in support of carbon capping strategies. The IT sector, which comprises nearly 2 percent of carbon emissions, has its own reasons for support green computing efforts. With the proliferation of massive datacenters over the last decade, sustainable computer system design and the adoption of renewable energy sources are essential.
A team of scientists from the Intelligent Design of Efficient Architectures Laboratory (IDEAL) at the University of Florida is working to develop more energy-efficient power supply schemes. In a recent paper, the team observes that the power consumption of big datacenters is skyrocketing. The energy demand for Google's servers has gone up 20-fold. Even if companies chose to ignore the profound environmental consequences of their carbon footprint, there is also the economic impact to consider, i.e., massive IT energy bills are not good for the bottom line.
According to a McKensey Quarterly report, the annual CO2 emissions of computing systems will research 1.54 metric gigatons within eight years. This sets a trajectory whereby the IT industry will be one of the biggest greenhouse gas emitters by 2020. As these data points stack up, so does the interest in green technologies. Renewable energy powered datacenters are gaining support from industry and academia because it's good for the environment and the bottom line.
Follow the Load
The IDEAL research team is investigating the use of cross-layer power management schemes as a method of enabling future sustainable high performance computing platforms. In a recent research paper, the team proposes the use of onsite distributed generation (DG) techniques to provide "premium clean energy to the computing load."
Onsite distributed generation is an emerging trend in the IT industry, according to the researchers, one that allows datacenters to utilize onsite green energy efficiently. The method they've developed, referred to as datacenter power demand shaping (PDS), takes advantage of a previously untapped power supply feature. It uses the load following capabilities of DG systems to avoid the high performance penalty issue incurred during supply tracking.
They write: "The novelty of PDS is two-fold. First, PDS intelligently trims data center load power and enables DG systems to follow the power demand efficiently. Second, PDS features two adaptive load tuning schemes that could boost data center performance and enable near-oracle operation during power demand trimming process. As a cross-layer power optimization scheme, our power management module resides between front-end distributed generation and back-end computing facilities to provide a coordinated tuning between the supply and load."
The design achieves near-oracle performance using a variety of green energy resources, according to the researchers. The term oracle represents the ideal energy balance scenario where fluctuating datacenter power demand is satisfied at all times with no performance overhead.
The design has a fundamental difference. Traditional systems force the IT load to track the variable power budget. In the team's approach, supply and demand are flipped: the renewable energy supply follows datacenter power demand. So there's very little wasted energy. In order to implement this approach, the team leveraged distributed generation with smart grid technology.
Distributed generation (DG) puts the power supply, in the form of small, modular electric generators, near the point of use. DG has been gaining ground as an alternative source of power for IT industry. The technology has garnered support from the U.S. Environmental Protection Agency (EPA), which stated that using DG in the datacenter has significant potential benefits, including great energy savings, a lower environmental impact, as well as high power reliability.
Using real-world datacenter traces and industry data, the team's solution comes within 1.2 percent performance of an ideal oracle. This reflects an improvement of about 37% compared to traditional supply-tracking based designs.
The bottom line: for a 10 MW datacenter, the technique would keep 100 metric tons of carbon emissions out of the atmosphere each year.
The group concludes:
"Overall, PDS could achieve 98.8% performance of an ideal oracle design and outperform existing supplytracking based approach by 37%. In addition, our design reduces electrical stress on batteries and could improve battery lifetime by up to 26%. The unique feature of load following based design makes this work a step forward toward the goal of incorporating clean energy resources into IT systems. We expect that our work could provide valuable guidelines for data center designers in the green computing and smart generation era."