A Multiprocessor Architecture for Substructure Search

The most time-consuming step in substructure search is the atom-by-atom match (ABAM). This is the reason why all commercially available search systems try to reduce the size of the search set by a preceding ‘pre-screening step’. The high CPU load imposed by ABAM is especially true when stereo information (configuration and conformation) is included in the matching algorithm. The need for an unusually large number of operations compared to the amount of data (i.e., 2 connection tables) on which it is operating is typical for an ABAM algorithm. This is in strong contrast to common numerical problems. This unusual requirement led to the construction of a search machine (TOPFIT) consisting of many independent parallel processing units and an administrating master processor which distributes the search data to the individual CPU’s and collects their answers consisting of one bit: does or does not fit.