[cryptography/bloomfilter] generic Hasher and optimizations #2729
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
Deploying with
|
| Status | Name | Latest Commit | Updated (UTC) |
|---|---|---|---|
| ✅ Deployment successful! View logs |
commonware-mcp | eb675f6 | Jan 19 2026, 08:18 PM |
Collaborator
Author
|
andresilva
commented
Jan 7, 2026
cryptography/src/bloomfilter/mod.rs
Outdated
| /// | ||
| /// The number of bits will be rounded up to the next power of 2. | ||
| pub fn new(hashers: NonZeroU8, bits: NonZeroUsize) -> Self { | ||
| let bits = bits.get().next_power_of_two(); |
Collaborator
Author
There was a problem hiding this comment.
Decided to just round up to the next power of two rather than crashing here. Let me know if you prefer asserting.
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
Deploying monorepo with
|
| Latest commit: |
eb675f6
|
| Status: | ✅ Deploy successful! |
| Preview URL: | https://db94ad89.monorepo-eu0.pages.dev |
| Branch Preview URL: | https://andre-bloomfilter-improvemen.monorepo-eu0.pages.dev |
andresilva
added
the
breaking-format
andresilva
force-pushed
the
andre/bloomfilter-improvements
branch
from
0ebf814 to
5589649
Compare
Collaborator
Author
|
bugbot run |
cryptography/src/bloomfilter/mod.rs
Outdated
| // Extract two 64-bit hash values from the SHA256 digest of the item | ||
| let digest = Sha256::hash(item); | ||
| let h1 = u64::from_be_bytes(digest[0..8].try_into().unwrap()); | ||
| let h2 = u64::from_be_bytes(digest[8..16].try_into().unwrap()); |
Contributor
There was a problem hiding this comment.
What was the rationale for u64 instead? I opted for u128 for security.
Collaborator
Author
There was a problem hiding this comment.
u64 arithmetic is faster than u128. My rationale was that bloom filters do not rely on collision resistance or large hash domains. All entropy beyond log2(m) bits is discarded during indexing, so using u128 does not improve security or false-positive behavior. The cryptographic hash should already provide more than enough uniformity.
andresilva
commented
Jan 8, 2026
cryptography/src/bloomfilter/mod.rs
Outdated
| let digest = Sha256::hash(item); | ||
| let h1 = u64::from_be_bytes(digest[0..8].try_into().unwrap()); | ||
| let mut h2 = u64::from_be_bytes(digest[8..16].try_into().unwrap()); | ||
| h2 |= 1; // make sure h2 is non-zero |
Collaborator
Author
There was a problem hiding this comment.
You have to be the unluckiest person in the universe for this to happen, but the Kirsch-Mitzenmacher optimization assumes that h2 is non-zero.
andresilva
changed the title
andresilva
changed the title
Collaborator
Author
|
I updated the benchmarks, here's a comparison of the impact of the optimizations: The optimizations have a big impact for small inputs size, for larger inputs computing the hash will dominate. Either way, this is consistently faster on all benchmarks. |
| /// | ||
| /// - **Performance**: Hash function performance varies with the size of items inserted | ||
| /// and queried. [Sha256] is faster for smaller items (up to ~2KB), while | ||
| /// [Blake3](crate::blake3::Blake3) is faster for larger items (4KB+). |
Contributor
There was a problem hiding this comment.
When adding Blake3, we found it was faster for both?
Collaborator
Author
There was a problem hiding this comment.
This is based on the benchmarks I added for bloom filter that run with both hash functions.
|
|
||
| impl<H: Hasher> BloomFilter<H> { | ||
| /// Compile-time assertion that the digest is at least 16 bytes. | ||
| const _ASSERT_DIGEST_AT_LEAST_16_BYTES: () = assert!( |
Contributor
There was a problem hiding this comment.
Wonder if this is overkill
Collaborator
Author
There was a problem hiding this comment.
If this is not true then we'll panic:
let h1 = u64::from_be_bytes(digest[0..8].try_into().unwrap());
let mut h2 = u64::from_be_bytes(digest[8..16].try_into().unwrap());
IMO makes sense to verify at compile-time.
Codecov Report❌ Patch coverage is
@@ Coverage Diff @@
## main #2729 +/- ##
==========================================
+ Coverage 92.24% 92.25% +0.01%
==========================================
Files 376 376
Lines 119098 119408 +310
==========================================
+ Hits 109856 110165 +309
- Misses 9242 9243 +1
... and 5 files with indirect coverage changes Continue to review full report in Codecov by Sentry.
🚀 New features to boost your workflow:
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
This PR makes the
BloomFiltergeneric overHasher, defaulting to Sha256.To support practical bloom filter sizing, a new
with_rateconstructor computes the optimal number of bits and hash functions for a given expected number of items and target false positive rate. Internally, this usesBigRationalto ensure deterministic results across platforms, important for consensus. This is the recommended way to construct bloom filters in most cases, as it automatically balances space efficiency against the desired accuracy.The index calculation has been optimized to use 64-bit arithmetic instead of 128-bit, and the bit count is now always rounded up to the next power of two. This allows using a bitmask instead of modulo for index calculation, which is faster.
Benchmarks were added to measure that the optimizations above work as expected. They compare Sha256 and Blake3 performance across different item sizes (32, 2048, 4096 bytes) and false positive rates (10%, 0.1%).