An Alternative Method to Remove Duplicate Tuples

The problem of performing database operations on parallel architectures has received much attention, both as applied and theoretical areas of research. Much of the attention has been focused on performing these operations on distributed-memory architectures, for example, a hyper-cube. Algorithms that perform, in particular, relational database operations on a hypercube typically exploit the hypercube's unique interconnectivity to not only process the relational operators eeciently but also perform dynamic load balancing. Certain relational operators (e.g., projection and union) can produce interim relations that contain duplicate tuples. As a result, an algorithm for a relational database system must address the issue of removing duplicate tuples from these interim relations. The algorithms accomplish this by compacting the relation into hypercubes of smaller and smaller dimensions. We present an alternative method for removing duplicate tuples from a relation that is distributed over a hypercube by using the embedded ring found in every hypercube. Through theoretical analysis of the algorithm and empirical observation, we demonstrate that using the ring to remove the duplicate tuples is signiicantly more eecient than using the hypercube.