Systems executing big-data workloads commonly move large volumes of data
through the memory and storage hierarchy, only to perform a modest amount of
computation on each data element. As a result, performance and energy-
efficiency of these systems is increasingly limited by the high cost of data
movement. To alleviate this bottleneck and leverage the high bandwidths within
memory and storage devices, there is a growing interest in abstractions and
technologies to move computation closer to the data.