Efficient Memory Storage for Concurrent Processing

			Doreen L. Galli, Ph.D.  


 In a parallel computing environment, we consider conflict-free access
 to constant-perimeter rectangular subarrays, using a natural
 formulation in terms of Latin squares.  For parallel matrix
 computations, there are frequently portions of data, referred to as
 templates, that one desires to be stored and retrieved in such a way
 that one can operate on the data simultaneously with each processor
 working on one element of the template. If more than one processor
 attempts to retrieve data from a single memory module during the same
 memory cycle, there is a memory conflict. When the array is stored to
 allow the desired templates to be accessible from the memory modules
 without memory conflicts, it is conflict-free for that set of
 templates. In addition to the results for a parallel environment, the
 methods and approaches have implications and hold promise in a
 distributed environment. The talk will conclude with a brief survey
 of a selection of Dr. Galli's other research projects.