Expand description
A split-block Bloom filter: 256-bit blocks, one block per key, one bit in each of the block’s eight 32-bit words. A probe touches one cache line. At 10 bits per key the false-positive rate is about 1 percent.
Keys are hashed with SHA-256; the first eight bytes, little-endian, pick
the block (high half) and the bits (low half). The bit of word i is
(low * SALT[i]) >> 27, the Parquet and Impala construction.
Structs§
- Bloom
- The filter over a set of keys.
Constants§
- BITS_
PER_ KEY - Bits per key the writer builds filters with.