Enormous histograms #53
Description
Cluster distributed histograms many 10s to 100s of GB in size with billions of bins are possible:
However, at present they are a little slow and could be improved with auxiliary data shuffling services in the filling step.
- What is the actual extent of uses cases for histograms this large?
- How do we make a good user interface to "virtual" histograms that are only ever rendered fully in memory back at the client machine?
- GPUs? GPU (CUDA)-kernel for end-user analysis #49