Performance improvements via transposing#38
Draft
kwodzicki wants to merge 3 commits into
Draft
Conversation
In Python/NumPy, iteration is 'fastest' over the last dimension of an array because data in this dimension are stored sequentially in RAM. In previous versions of this package, data were expected to be ordered such that dimension were [time, y, x], or some variant with time being the left-most, and slowest, dimension. This update flips how data are processed and assumes that the right-most dimension is time. This enables faster compute as the time series of data for a single location are now sequential in RAM, leading to speedups in both serial and parallel performance. I have also reduced the number of processes that run in the parallel case by only iterating over the last dimension. In the previous version, iteration occured over two dimension, leading to lots of 'small' chunks of data being passed back and forth to the processing Pool. This communication can be slow and incurs some overhead. To reduce this, we now pass 'large' chunks of data to a single process and use the map_over_location() static method to iterate down to the time dimension. Testing indicated a 50% reduction in processing time, with only a slight increase in read time (10-20s) when transposing the data on read via Xarray.
Modified the dimensionality requirment to check that data are at least 1-D (i.e., array.ndims > 0). Updates to code for speed improvements made things more flexible with dimensionality, so do not have to worry as long as time is last/right most dimension and all other dimensions match across the arrays, everything should work as expected. Closes ecmwf-projects#22
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
In Python/NumPy, iteration is 'fastest' over the last dimension of an array because data in this dimension are stored sequentially in RAM.
NOTE: This is a breaking change!!!
In previous versions of this package, data were expected to be ordered such that dimension were [time, y, x], or some variant with time being the left-most, and slowest, dimension.
This update flips how data are processed and assumes that the right-most dimension is time. This enables faster compute as the time series of data for a single location are now sequential in RAM, leading to speedups in both serial and parallel performance.
I have also reduced the number of processes that run in the parallel case by only iterating over the last dimension. In the previous version, iteration occured over two dimension, leading to lots of 'small' chunks of data being passed back and forth to the processing Pool. This communication can be slow and incurs some overhead.
To reduce this, we now pass 'large' chunks of data to a single process and use the map_over_location() static method to iterate down to the time dimension.
Testing indicated a 50% reduction in processing time, with only a slight increase in read time (10-20s) when transposing the data on read via Xarray.
Closes #37