Documentation Benchmarks
TUNING / MEASURE BEFORE CHANGING

Read caching, merge costs, and storage sizing

Compare a representative workload against the default configuration, then use read latency, merge I/O, file shape, or payload cost to choose an experiment.

01 / READ WORK

Read-path caching

Separate block retention, deserialized-record caches, sequential lookup hints, and scan prefetch.

Match caches to reads
02 / MERGE WORK

Write amplification

Understand reuse eligibility and observe unchanged data files across a localized update.

Measure rewrite work
03 / FILE SHAPE

Disk segment tuning

Distinguish writable capacity, disk-layer rollover, part targets, block size, and sparse density.

Segment and part sizing
04 / PAYLOAD COST

Large values

Translate record counts into bytes, evaluate value ownership, and compare direct payloads with references.

Budget payload work
Make each experiment comparable
  1. Fix the dataset, key distribution, payload sizes, concurrency, and query bounds.
  2. Separate cold and warm reads, and state the engine-cache and OS-cache conditions.
  3. Change one control or coupled target range, and record the exact options.
  4. Compare latency distributions, throughput, allocations, retained memory, and I/O through a representative maintenance cycle.
Storage fundamentals