Every field of science and commerce now relies on computers and their capability to process data and information—fast.
Moore’s law enabled doubling the number of transistors that can be put on a chip every 18 months.
The ever-growing performance of computers is due to two main factors: our ability to shrink electronic circuits to smaller and smaller dimensions, and architectural innovation that improves how circuit elements interact with each other to perform computations. The former is commonly known as Moore’s law. Moore’s law enabled doubling the number of transistors that can be put on a chip every 18 months. This success continued for several decades.
When it became clear this path can’t be sustained, processor manufacturers shifted their focus to putting more processors on a single chip. Unfortunately, these processors cannot be programmed using the traditional software technique of writing sequential programs. Instead, programmers have to develop parallel programs. Parallel programs are difficult to write, debug, and maintain.
Since his CAREER award in 2003, Soner Onder has worked on alternative methods of building and programming processors. Contrary to existing execution paradigms, called Von Neumann architectures, Onder developed an alternative execution paradigm called Demand-Driven Execution. In this form of processor architecture and programming model, the execution of programs proceeds from the outputs of a program toward its inputs, and in the process, the machine automatically extracts all the available parallelism in the program and maps the extracted parallel operations to multiple processing elements. Unlike so-called multi-core computers, his approach does not require parallel programming. Instead, the execution paradigm itself automatically executes a given sequential program in parallel.
Onder’s most recent NSF grant is through the Exploiting Parallelism and Scalability program, which, according to NSF, “supports groundbreaking research leading to a new era of
parallel computing.”
Michigan Tech is the lead institution in this four-year project totaling $875,000. The project will study demand-driven computing to meet existing and emerging workload demands. Onder is working with four PhD students and colleagues from Florida State University to establish demand-driven computing as a viable approach to build faster and more power efficient computers.