The paper describes a slice-based data-allocation strategy for tree-based topologies. The approach is supported by a theoretical analysis demonstrating the optimality of the data-distribution procedure. According to its basic principle of operation, data is split in such a way that at run time no processor ever stands idle. The benefits of this approach are quite important in several practical applications, including high-dimensional data processing and neural network modeling. Experimental results obtained from a noise-like coding model of associative memory confirm the validity of the overall approach.
Index Terms:
resource allocation; optimal data allocation; processor-tree architectures; slice-based data-allocation strategy; theoretical analysis; data-distribution procedure; high-dimensional data processing; neural network modeling; noise-like coding model; associative memory
Citation:
F. Ancona, R. Zunino, "Optimal data allocation for processor-tree architectures," euromicro, pp.351, 23rd EUROMICRO Conference '97 New Frontiers of Information Technology, 1997