vb-17: fix cach eviction discrepancies#19
Open
nvzm123 wants to merge 9 commits into
Open
Conversation
Author
|
@nvzm123 i would really like us to put effort into the cuVS-lucene backend for cuVS-bench, rather than continue to improve these scripts. Is that something you could take on? It's possible Claude might be able to assist with this. |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
We weren't prewarming the data before running search which led to huge search-time discrepancies. This PR adds prewarming after we built our index and potentially evicted from our cache, we reread vectors from disk to prewarm cache.
E.g. with an index size of 10M 1536d 32fp (~58 GB) and a dynamic Java Heap that spikes above 70GB, depending on our Host RAM size, we were having to always spend the price to reload data from disk that was evicted when running g5.8x (128 GB Host RAM) and frequently also seeing the same on r8gd.8x (256 GB RAM). With this fix, those discrepancies disappear and we see the results we would have expected given the same benchmarks at 1M (where the Java Heap never spiked so large as to evict cached data).