Smarter data movement for faster computing
How can we make our computers faster? The answer, according to David Black Schaffer’s new European Research Council Starting Grant project, lies in being cleverer about how we move and track data. “The key is to share information between the hardware and the software such that we only move data when it is needed and to the places where it will be used. If we can do that, we will save time and energy.”
As a computer architect at the Department of Information Technology, David Black-Schaffer’s research focuses on developing approaches to increase computer performance. His major challenge is that today’s computer systems are power-limited, meaning that any increases in performance must come from more power-efficient designs. This is true across the board, from mobile devices where battery size is fixed to datacenters, where cooling is limited, says David Black-Schaffer.
“The energy used to move data inside the computer processor is greater than that used to actually compute our answer. The reason is simply that today’s systems use the brute-force approach of searching through the memory system to find and retrieve data. If we avoid searching and instead keep track of where the data is directly, we can access the data more quickly and more efficiently.”
While knowing where the data is located allows us to access it more efficiently, David Black Schaffer points out a greater challenge: learning where to put the data in the first place. To accomplish this, his 2016 ERC Starting Grant project will investigate how to integrate information from both the hardware and the software for smarter data placement and movement. The grant of 1.5M Euros over five years will enable him to hire a group of doctoral students and post-docs to tackle this challenge.
Patented techniques
The ERC-project builds on several foundational techniques that he has developed together with Professor Erik Hagersten and doctoral student Andreas Sembrant.
“We have developed, published, and patented several radical new techniques for efficiently tracking and accessing data. And through our startup, we have been working with companies to evaluate them and get feedback,” says David Black-Schaffer.
“But the biggest challenge is that industry is ten percent innovation and ninety percent validation, while academia is the other way around. Proposing radical new designs to industry is always a challenge, and we have been learning a lot about what they require to adopt these new approaches. However, ERC projects are about high-risk as well as high-reward research, so it is important to balance the two.”
The indispensable role of computers in science and society, for everything from weather prediction to drug development, and even entertainment, makes it critical to keep increasing computing performance, says David Black-Schaffer.
“Many people wonder ‘Why do I need a faster computer?’ Well, you may not for what you do today, but if we had computers that were ten times faster we could make our weather predictions twice as accurate, which would save billions of euros in crop losses and storm damage every year. If we are wildly successful in this project, we can have a dramatic impact on pretty much all aspects of society.”
Chronicle: Delivering efficiency with heterogeneous computation
More on David Black-Schaffer’s research
Anneli Björkman