• map
• multi-core
• gpgpu
• parallel patterns
• cost models
• reduce

The purpose of this thesis is to present some possible implementations and introduce cost models for evaluating the performance of data-parallel computations (map, reduce) that use both the CPU and GPU in order to maximize resource usage. We will then use these cost models to determine the best input data split between the two computational units that optimizes performance and study its precision on a few instances of the aforementioned data-parallel paradigms.