Bloom Filters

Bloom filters are probabilistic data structures that test whether a key might be in an SSTable, enabling fast negative lookups.

How It Works

1. When an SSTable is created, every key is hashed into the bloom filter
2. On a point lookup, the bloom filter is checked before reading any data blocks
3. Negative result: The key is definitely NOT in this SSTable → skip it
4. Positive result: The key MIGHT be in this SSTable → read the data blocks

False Positive Rate

With the default 10 bits per key:

• False positive rate: ~1%
• Memory cost: ~1.25 bytes per key

Bits per Key	False Positive Rate	Memory per 1M Keys
4	~5.6%	0.5 MB
8	~2.2%	1.0 MB
10	~1.0%	1.25 MB
14	~0.2%	1.75 MB
20	~0.01%	2.5 MB

Impact on Read Performance

Bloom filters are critical for read performance:

• Without bloom filters: Every SSTable must be checked for every read
• With bloom filters: ~99% of SSTables are skipped per read

Configuration

Parameter	Default	Description
bloom_bits_per_key	10	Bits per key (minimum: 4)

Tip

For read-heavy workloads, increase bloom_bits_per_key to 14 or higher to reduce false positives.