Parallel et Tree Hashing Référence
Référence pour parallelizable hachage constructions incluant BLAKE3, KangarooTwelve, et ParallelHash.
Parallelization Strategies
Tree hashing splits input into chunks, hashes each independently, then combines results in a tree structure. This enables multi-core parallelism and incremental updates. BLAKE3 uses a binary Merkle tree of chunk hashes with a compression function. KangarooTwelve extends Keccak with a tree of sponge calls for parallelism. NIST ParallelHash (SP 800-185) provides a standardized parallel SHA-3 variant.
- tree hashing
- Independent chunks combined
- BLAKE3
- Binary tree of chunks
- KangarooTwelve
- Keccak-based tree
multi-corelinear speedupXOFPerformance and Tradeoffs
Parallel hashes achieve near-linear speedup with core count for large inputs. For small inputs, the tree overhead makes them slower than sequential hashes. Incremental hashing allows updating a portion of input without rehashing everything. The tree structure also enables proof-of-inclusion for individual chunks. Memory bandwidth often becomes the bottleneck beyond a certain core count.
- large input
- Near-linear speedup
- small input
- Tree overhead, slower
- incremental
- Update without full rehash
multi-coreuse sequentialedit large files