CSAIL principal investigator Charles Leiserson - a professor in MIT’s department of Electrical Engineering and Computer Science (EECS) - recently co-authored a paper published by the Computing Research Association as part of the first edition of their “Quadrennial Papers,” published every four years.
In the paper, Leiserson and his colleagues argue that the end of Moore's Law will require the U.S. to put a more concerted national focus on Software Performance Engineering (SPE) in order to maintain its status as a global technology leader.
Moore’s Law has led to transformational changes in technology, with the miniaturization of semiconductors leading to exponentially cheaper, more efficient computer hardware. But Leiserson and team say that future performance will have to come instead from SPE — that is, optimizing code to run faster and use fewer resources.
Unlike hardware improvements that benefit all applications, SPE requires optimization for specific applications. "Although shared libraries and platforms can be performance-engineered to run faster, achieving high-performance applications typically requires custom engineering of the applications themselves," Leiserson says.
One concrete issue in being able to shift to prioritizing SPE is that very few software engineers are currently proficient in the field, highlighting a critical skills gap. Comprehensive SPE education and training programs are needed at all levels, from undergraduate to professional development. Leiserson says that research universities should be encouraged to develop and offer SPE curricula within their full-time degree programs, and also makes the case that substantial government funding will be needed for the U.S. to remain competitive - especially in light of analogous recent legislation in the form of the hardware-focused CHIPS Act.
"A complementary effort focused on SPE will ensure that the CHIPS investment is not undermined by producing ostensibly high-performing hardware that cannot be effectively programmed,” Leiserson says.
On top of that, the authors emphasize the importance of more research in SPE techniques, programming languages that support performance, and instrumentation for performance analysis.
Existing software-productivity tools are also outdated and require significant updates to enable efficient SPE.
"Without immediate investment, we risk jeopardizing the benefits of an economically strong and free society for ourselves and future generations,” Leiserson says.
CRA’s Quadrennial Papers are a series of white papers published every four years that delve into key areas and issues within computing research that have the potential to address national priorities. They aim to present a comprehensive overview of the computing research field, outlining prospective research directions, identifying challenges, and offering recommendations for policymakers and the broader computing research community.
