Abstract:

Multidimensional data are processed and accessed in the linear form. Adjacent nodes in 2dimension or higher space are mapped to nonadjacent addresses in 1dimensional storage hardware. These variant address distances lead to variant access latencies. We propose a storage mapping algorithm based on ZOrdering function, and several indicators of this algorithm are presented and calculated, which are compared with the ordinary rowmajor ordering mapping method. The proposed algorithm manifests a stronger locality to aggregate the adjacent nodes in a small linear scope. An increscent prefetching space in a fixed cache space maximizes this strength, therefore, the cache hit rate is increased and its system performance is improved.

Key words: multidimensional data;storage mapping;cache;prefetching;hit rate