Increasingly, jobs rely on the ability to use computers to interpret, understand, and trust data. For example, my students and I have worked with ornithologists who cannot understand the representations of their bird sightings, civil engineers who cannot easily use their own building data, finance experts who cannot trace money between companies and their subsidiaries, and an XML document company whose clients cannot understand data that appears outside of their reports. In each case, the data users have been hampered because their data is exceedingly difficult to understand and trust, even though the users are experts in their fields. One reason for this difficulty is that the organization of the data is often designed for computers, not for people (i.e., for storage, not accessibility). Another reason is that data often come from different sources, leaving users with the challenge of integrating data that they neither understand nor trust.

As a researcher, I am fascinated by the challenge of advancing the high-level foundations of computer software (programming models, compilers, and runtimes) to productively exploit the latest advances in computing systems. While there has been a long tradition of research in this area since the dawn of computing, the rapid evolution of hardware has continuously fueled a need for new software technologies as old approaches quickly become obsolete. Current explorations of new hardware directions that go beyond Moore’s law have further amplified the motivation for this research direction.

What value should a memory read return? The answer to this simple question is surprisingly complex for modern systems running parallel software. The memory consistency model, which governs this answer, is a fundamental part of the hardware-software interface, but has been one of the most challenging and contentious areas in parallel hardware and software specification. […]