Libraries
ArrayRoute
Read chunked scientific arrays within an explicit memory budget.
View source on GitHubAbout ArrayRoute
ArrayRoute plans selections over chunked scientific arrays before reading them. Zarr, TensorStore and HDF5 backends share one read protocol, so large simulation fields can be streamed tile by tile with bounded decoded memory and measured read statistics. It is independent of Frontier and needs no account, scheduler or cloud credentials.
- Plan a selection in time proportional to array rank, even for petabyte logical shapes.
- Stream selections too large to materialize as bounded tiles.
- Compute mergeable statistics and compare reader policies with reproducible benchmarks.
Before you use it
Experimental. NumPy is the only core runtime dependency; storage engines are optional extras.
See how it connects to Frontier