Avoid pinned memory for shuffle host buffers #1909
Merged
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Outdated suggestions cannot be applied.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.
When using legacy Spark shuffle the buffers are first collected into host buffers that are later concatenated into one monolithic host buffer by
GpuShuffleCoalesceExec
that is sent to the GPU. The intermediate buffers do not need to be allocated with pinned memory because they are not sent directly to the device. This reduces the load on the pinned memory pool. Because the pinned memory pool allocator currently does not perform well with many allocations, see rapidsai/cudf#7553, this change shows noticeable speedup on regular queries with the default 200 partitions.