Research Statement

The computer industry has thrived upon decades of exponential growth in hardware and software capabilities. As hardware becomes more powerful and cost-efficient (primarily facilitated by Moore’s Law), software expands to utilize the additional capacity. This expansion of software in turn drives the development of better hardware. Underlying this synergistic evolution is an increasing human ability—assisted by better programming models and software tools—to build bigger and more complex systems. In the future too, increases in human productivity are going to be key to sustaining this model of evolution and expansion of the computer industry. Recent industry trends, however, have cast a shadow on programmer productivity gains in software development. The shift to chip multiproces-sors (CMP) as the vehicles for general-purpose computing has signalled a move to writing parallel software. However, thus far, parallel software has been hard to write and bug-prone. This problem is exacerbated by the lack of efficient software tools to facilitate parallel programming. Unless we in-novate in parallel programming models and software tools, we are likely to run into a productivity wall that would halt the evolutionary cycle driving our industry. My research focusses on providing hardware support to ease soft-ware programming and thus improving programmer productivity. This support can be in the form of either facilitating better programming models like Transactional Memory (TM) [12, 13] or more efficient software tools like dynamic atomicity checkers [14]. In my dissertation research, I provide a foundation for a broad class of such proposals, known as supervised systems, along with exploring applications of supervised systems