Duncan Moss (University of Sydney researcher and PhD candidate in the School of Electrical and Information Technologies)
Algorithms designed by a University of Sydney student may reduce the high electricity consumption of Datacentres and Cloud facilities to not only deliver information faster, but also to better the environment.
As a result, Duncan Moss, a PhD candidate in the School of Electrical and Information Technologies, has been invited to undertake a yearlong research internship with Intel at its Datacentre Cloud Platform Group at Hillsboro in the US.
Specifically, Moss will join Intel’s team to investigate better methods for using new chips to be released in 2017 which combine central processing unit (CPUs) and field-programmable gate array (FPGAs).
“This is an amazing opportunity for me. The particular FPGA algorithm that I will be able to develop will allow Datacentres to run with less power but just as efficiently. FPGAs are specialty chips used to speed up certain kinds of computer workloads such as machine learning. The FPGA chip has also been around for a while but is difficult to work with, however Intel’s new technology will make it mainstream,” he said.
“It’s like being given two buckets of Lego blocks. The CPU bucket only has one type of Lego block so it can only be made into one thing, but the FPGA bucket has all kinds of blocks – small, large, rectangle or square – an assortment that can be changed into any desired configuration. We can change the FPGA part to suit our needs but the CPU can’t be altered. Also FPGAs are far more energy efficient than CPUs on a performance-per-watt basis.”
Moss added that Datacentres are avid consumers of power and if allowed to continue to expand in their current form, could become a major consumer of the world’s energy supply.
Philip Leong, Moss’s’ Ph.D. supervisor agreed. He said, “Greenpeace noted that if the Cloud were a country, it would have the fifth largest electricity demand in the world. Moss’s work is extremely timely as it will utilise Intel’s technology to dramatically improve the speed and power consumption of these Datacentres.”
