• aggregate operations
• BatchedFAT
• CUDA
• data parallelism
• FastFlow
• FlatFAT
• gpu
• graphical processing unit
• sliding windows
• stream processing
• streaming analytics
• WindFlow

Sliding windows are characterized by the fact that consecutive windows share elements. When dealing with aggregate operations this property can be used to improve the number of
operations necessary to compute the aggregate value of the window. There are many works that tackle this problem by designing a data structure that keeps track of partial aggregates of the shared elements that are used to compute the final aggregate value. We would like to explore the role GPUs can play in improving performance of aggregate operations on sliding windows. The work starts by giving an overview about streaming analitycs, window processing, parallelism in streaming applications, WindFlow and how GPUs work. Then proceeds to give a detailed analysis of FlatFAT, one of the data structure designed to exploit overlaps between sliding windows, and its implementation in the WindFlow library. Then it shows a parallelization of this data structure that can be managed by the GPU, called BatchedFAT, and its relative implementation inside the library. Finally it presents a comparison between the basic approach of WindFlow, FlatFAT and BatchedFAT.