perf: add prefetching for aggregate multi group by [WIP] #19520
Conversation
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
github-actions
bot
added
common
Member
Author
|
@alamb, @Dandandan you might be interested in this (not ready for review, just the general idea) |
rluvaton
commented
Dec 28, 2025
Comment on lines
+653
to
+655
| pub agg_prefetch_elements: usize, default = 1 | ||
| pub agg_prefetch_locality: usize, default = 3 | ||
| pub agg_prefetch_read: bool, default = false |
Member
Author
There was a problem hiding this comment.
Added config to allow for easy tinkering while prototyping
rluvaton
commented
Dec 28, 2025
Comment on lines
+550
to
+562
| match self.agg_prefetch_elements { | ||
| 0 => self.collect_vectorized_process_context_with_prefetch::<0>(batch_hashes, groups), | ||
| 1 => self.collect_vectorized_process_context_with_prefetch::<1>(batch_hashes, groups), | ||
| 2 => self.collect_vectorized_process_context_with_prefetch::<2>(batch_hashes, groups), | ||
| 3 => self.collect_vectorized_process_context_with_prefetch::<3>(batch_hashes, groups), | ||
| 4 => self.collect_vectorized_process_context_with_prefetch::<4>(batch_hashes, groups), | ||
| 5 => self.collect_vectorized_process_context_with_prefetch::<5>(batch_hashes, groups), | ||
| 6 => self.collect_vectorized_process_context_with_prefetch::<6>(batch_hashes, groups), | ||
| 7 => self.collect_vectorized_process_context_with_prefetch::<7>(batch_hashes, groups), | ||
| 8 => self.collect_vectorized_process_context_with_prefetch::<8>(batch_hashes, groups), | ||
| _ => self.collect_vectorized_process_context_with_prefetch::<8>(batch_hashes, groups), | ||
| } | ||
| } |
Member
Author
There was a problem hiding this comment.
Added all of these to make sure when I test specific config I'm not affected by the condition, will not have this later
rluvaton
commented
Dec 28, 2025
| } | ||
| } | ||
|
|
||
| #[inline(always)] |
Member
Author
There was a problem hiding this comment.
Inline always to make sure my change for extracting to a function is not affecting the prev case
rluvaton
commented
Dec 28, 2025
Comment on lines
+613
to
+619
| for i in 1..=PREFETCH { | ||
| if READ { | ||
| self.map.prefetch_read::<LOCALITY>(batch_hashes[row + i]); | ||
| } else { | ||
| self.map.prefetch_write::<LOCALITY>(batch_hashes[row + i]); | ||
| } | ||
| } |
Member
Author
There was a problem hiding this comment.
I prefetch the same item multiple times but I should prefetch before the loop and then only prefetch the row + PREFETCH
Member
Author
|
run benchmarks |
|
🤖 |
Member
Author
|
Not sure how much the existing benchmarks will show improvement as they are not testing a lot of unique values |
|
🤖: Benchmark completed Details
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
Which issue does this PR close?
N/A
Rationale for this change
grouping on large amount of data is very slow
What changes are included in this PR?
used my fork of hashbrown with prefetching and added prefetching when map is large
Are these changes tested?
No
Are there any user-facing changes?
No
Related hashbrown issue for allowing prefetching:
My hashbrown changes:
my hashbrown changes are not very safe and probably have bugs but just to make it quick and dirty to see the benefit
Command:
datafusion-cli --command "select value from range(0,100000000) group by value, value + 1;"When the hashmap is smaller, the prefetching is unneeded as it should already fit in the cache
with the following config:
Without prefetch: 16.002 seconds
With prefetch: 10.176 seconds
Env
Machine:
c5.metalResults
Without prefetch
With prefetch