| OCR Text |
Show 270 adaptive and self-adjusting data structures [119], data and knowledge representation, and integration technique for different data representation, etc. Second, if a processor pattern can be designed that just matches this (m.init:num spatial entropy) data distribution pattern, a least amount of spatial data entropy would be generated between both structures. Therefore, for a given data distribution pattern with minimum spatial data entropy, a physical computing structure to a great extent determines the amount of spatial data entropy between both structures. In summary, it is the data distribution pattern, the processor distribution pattern (with processor granularity) and the degree of their matching that determine the amount of spatial data entropy in a computer. Reducing spatial data entropy increases data parallelism and therefore the computation efficiency. Temporal data entropy can be reduced or removed by performing control instructions sequentially (which is very inefficient) , or by parallelizing the computing sequence that is executed in a strictly sequential fashion. Similarly, the second strategy can be implemented in two steps. First, there is always one data processing sequence that produces the minimum amount of inherent temporal data potential. The study of algorithm control and data dependency as well as machine instruction execution, for example, offers insight to determine efficient parallel data processing sequence of the given problem structure. During the second step, one should identify the threads of control of a computing structure that could operate independently and parallelize those sequential processes whenever possible to match that (minimum temporal entropy) data processing sequence. The primary problem is the difficulty of identifying and synchronizing these independent threads of control. For a given problem structure that possesses a natural data processing sequence, the algorithm control of a computing structure dominates the amount of temporal entropy between both structures. In summary, it is the inherent data processing |
