slimdb · research

SlimDB

SlimDB is a research key-value store that combines compact indexes and configurable merge layouts for workloads requiring prefix-group scans without suffix ordering.

Query contract: semi-sorted keys have a prefix and suffix; scans retrieve a prefix group with no required suffix order. The three-level SSTable index compresses prefix and suffix information separately. Its entropy-coded trie locates candidates, not membership. [1]

Lookup scope: the multi-level cuckoo filter records fingerprints and sub-level identifiers. A secondary table keeps full keys for collisions and is checked first. Its one-candidate argument is not a whole-engine I/O guarantee: uncached block indexes add reads, and blind writes can exhaust the secondary table's budget and lose the bound. [1]

Trade-off: the paper reports slower in-memory index lookup and slower prefix scans in its stepped-merge configuration despite memory savings. These are historical, workload-specific results, not a current RocksDB ranking. [1]

Illustration: a group containing suffixes 7, 1 and 4 may be enumerated in that order. Taking its first member at least 3 returns 7, whereas the smallest qualifying suffix is 4. An unordered group result needs additional comparison or sorting before it can answer that ordered query. This is an original contract example, not an engine trace.

Contrast: RocksDB prefix filtering skips runs without a selected prefix within its comparator-based iterator contract; it does not make the suffix order irrelevant to applications. [2]

Relations

related

Log-structured merge-tree, SILT, Cuckoo filter, Prefix Bloom filter, Prefix extractor, Read amplification

Citations

1. Ren et al. SlimDB: A Space-Efficient Key-Value Storage Engine For Semi-Sorted Data. PVLDB 10(13), 2037-2048 (2017). DOI: 10.14778/3151106.3151108. (accessed 2026-11-14)
2. RocksDB project. Prefix Seek: prefix definition, comparator compatibility and iterator modes. (accessed 2026-11-14)

JSON: /data/terms/slimdb.json