A Reliable Parallel Backend Using Multiattribute Clustering and Select- Join Operator

An access method based upon muhi-afiribure clustering allows fhe database administrafor to define muhiply parlilioned relalions. For each attribute in the clusrering, we can consider the set of subrelations as a relarion view. Such a method has been implemented in SABRE. II relies on muhi-a~tribute digital hashing and a linearly growing directory. Using this method, we show that it is possible IO improve the multiprocessor hashing join algorilhms by a ratio of 3 to 5, with rhe same hardware configurarion. According IO our evaluation, the memory requirements are approximately the same as wirh the hashing algori#uns, and the common bus used for disk accesses does not salurate. Any configurarion can be linearly extended by adding or removing a disk or a processor, and reIiabilily is guaranteed by a simple management of multiple copies. In case of a disk breakdown, the conrinualion of operalion is possible wirh minimum loss of speed.