Professor Albert Zomaya, director of high performance computing at University of Sydney.
Newly developed algorithms may be the solution to reducing the increasing energy consumption and cost of high performance computing (HPC) networks, according to University of Sydney director of HPC, Professor Albert Zomaya.
Speaking to CIO Australia, he said that in the last few years the University has patented a “very sophisticated” algorithm which performs energy reduction.
"That algorithm deals with energy consumption by manipulating voltages at a processor level,” Zomaya said.
“We know that modern processors can operate at different voltage levels and by manipulating these voltages we are able to run a workload without compromising the execution time or the quality of service while at the same time reducing the energy consumption of the platform."
According to the professor, this platform could be a small or medium sized data centre which has hundreds of processors.
Some HPC systems use megawatts of electricity for operation and cooling, Zomaya said.
“On average, power bills for such systems can run in the millions per year.
“From the results we obtained from our extensive algorithm simulations we can see that, depending on the nature of the [HPC] application, the savings can run from five per cent to 35 per cent.”
On average, HPC network operators could save at least 15 to 20 per cent in energy bills if they used the algorithm.
While the algorithm has not been released for commercial use yet, the University of Sydney has implemented an extended version of the algorithm on a prototype data centre at its HPC division.
“We currently have two prototype data centres in the HPC centre as we’re running different hardware, metering and tools,” he said.
This is so university staff can gauge the exact energy consumption levels and experiment with different workloads to get a more complete picture of energy consumption profiles.
“The hope is that within the next six months we should be able to have some of these solutions implemented in the hardware and properly tuned to deal with different case studies,” Zomaya said. “After we do the prototyping here it will be nice to run this [algorithm] in a production environment to see how it is going to perform over a long period of time.”
Follow Hamish Barwick on Twitter: @HamishBarwick