PPS-a parallel partition sort algorithm for multiprocessor database systems

A new algorithm, parallel partition sort (PPS), is proposed which is an improved range partition sort algorithm to handle the problem of workload imbalance, especially partition imbalance and heterogeneous imbalance, in a shared-nothing multiprocessor database environment. The new algorithm partitions the key range into a number of range intervals, then a fast internal sorting method is applied on each range interval combined with a dynamic mechanism to handle different interval sizes. A dynamic mathematical model approach is used to balance the workload among processing nodes by estimating data distribution during the sorting process. Experimental results demonstrate that the new algorithm performs better than existing parallel range partition sorting algorithms in a shared-nothing database environment for a wide degree of skew