Expert: Software the next battlefield for datacentre efficiency

Scientists have developed a new low-cost algorithm that promises to cut datacentre power usage by up to 60 per cent with no changes to hardware.

The Energy Conscious Scheduling algorithm (ECS), which has been patented by University of Sydney’s Centre for Distributed and High Performance Computing researchers, Dr Young Choon Lee and Professor Albert Zomaya, works by mapping the tasks performed by computers and scaling their energy usage as required.

“People have been looking at energy issues for a long time but they’ve been looking at issues of energy consumption at the level of the circuits and transistors,” Zomaya said. “The idea here is to not only look at the energy consumption of the circuit or chip, but to go beyond that to the system level or the node level in a more holistic fashion.

“This algorithm would normally sit comfortably in the operating system or be part of the middleware that sits on top.”

Zomaya said the ECS would not require any hardware modifications and could be layered on top of an existing operating system with a minimum of disruption, however detailed preparation was required for each case.

“A lot of crafting goes into this kind of thing before the algorithm is ready because you have to really do a lot of analyses and simulations to make sure it is ready for a range of applications and not just one,” he said.

The ECS’s mathematic nature means it can work across a number of different platforms regardless of vendor. The researchers’ ultimate aim is to develop an autonomous program that can map the system and cut power usage by 10 – 60 per cent without human intervention.

Zomaya said rival algorithms would soon reach the market and that software would be the next big solution for datacentre energy efficiency.

“Over the next two to three years, we’ll see this energy-aware type of software or middleware become more commonplace and it’s going to be a part of the new generation of machine,” he claimed.

The ECS is set to be demonstrated in a prototype datacentre by the end of this year and released commercially by 2011.