Theory and Algorithms for Multicore Computing

University Of Texas At Austin, Austin TX

Investigators

Abstract

Chip multiprocessors (i.e., multicores) are rapidly emerging as the dominant platform for computing. This computing environment features parallel computing with shared-memory and a hierarchy of caches. This research project will develop a theory of multicore computing, which will include (but is not limited to) the following aspects: Developing and refining abstract theoretical models for multicore computing; designing and analyzing efficient algorithms for multicores for important problems, including cache-oblivious and processor-oblivious algorithms; designing schedulers for efficient execution of parallel algorithms on multicores; and developing lower bounds to delineate the inherent limitations of multicore computing. The outcome of the project will be a new set of methods for the efficient use of multicores. Since multicores are widely expected to be the dominant compute platform for the foreseeable future, these results are likely to have broad applicability to all fields that rely on compute-intensive tasks.