Comparison
Read amplificationvsWrite amplification
Read amplification
one point lookup turns into six file reads because the key could be in any of the levels the compactor has not merged yet.
How many physical reads it takes to answer one logical read. LSM engines trade write amplification down and read amplification up, mitigating it with bloom filters and compaction. It shows up as a p99 that is far worse than the median, because the tail is the requests that had to check every level.
Full entry →Write amplification
you wrote one kilobyte and the disk did fifteen kilobytes of work, because the row is in three indexes and compaction will rewrite it twice.
The ratio of physical bytes written to logical bytes you asked to store. It comes from index maintenance, write-ahead logging, page-level rewrites and background compaction. It is the reason adding "just one more index" degrades write latency, and the reason an SSD wears out faster than the raw data volume suggests.
Full entry →