Supercomputing group sets sights on 'green' cloud

Researchers compare Microsoft, Amazon clouds.

Sydney University researchers have engaged industry partners to improve the efficiency of the “next iteration” of supercomputers: enterprise data centres and the cloud.

According to Albert Zomaya, who chaired the university’s Centre for Distributed and High Performance Computing (CDHPC), the cloud computing industry was “still very much in need” of Green IT.

“Power is the biggest problem here,” Zomaya told iTnews, after receiving two IEEE awards for his contributions to parallel processing and scalable computing last month.

“The idea is how we use these [data centre] systems in a better way, and at the same time, not decrease performance.”

Zomaya said today’s virtualised data centres were “another dimension of high performance computing systems”, which also included multiprocessor-clusters and monolithic supercomputers.

Compared to on-premise applications, cloud applications were more susceptible to latency, so performance came with a higher degree of uncertainty.

Number crunching and the movement of large amounts of data cost energy, which was becoming increasingly expensive as power costs increased, Zomaya said.

CDHPC researchers hoped to either “tune” or replace cloud applications so they would perform most efficiently in various computing-as-a-service environments.

The effort had $443,169 in funding from the Australian Research Council and Microsoft, which also began providing the researchers with access to its Azure cloud platform this year.

Researchers have also experimented on the Amazon Web Services platform since early 2010, as well as using an internal, prototype cloud environment.

Zomaya hoped to discover how the platforms handled suites of applications, expecting there to be “different horses for difference courses”.

“One needs to understand which clouds are good for different applications,” he said, describing an arduous testing process that harked back to supercomputing application development.

“What we’re hoping to do is go back to our friends at Microsoft and hopefully that will start a dialogue [about improving the efficiency of Azure].”

In September 2009, CDHPC researchers patented an ‘Energy Conscious Scheduling (ECS)’ algorithm that they said could halve the energy consumption of processors in data centres at little or no cost.

Zomaya said the researchers were working on improving the efficiency of software, middleware, hardware and buildings, and “profiling” real-world environments so they can be compared.

He declined to comment on the efficiency and performance of commercial offerings, suggesting only that technologies developed by vendors on the Green 500 supercomputing list also applied to virtualised data centres.

The CDHPC was also in negotiations with an unnamed Australian bank about “greening its data centre environment”, he said.

“There are several players in the marketplace that are doing that; this energy business today is very, very timely.”