Work out compaction overhead instantly with clear inputs, formula shown and shareable results.
Compaction consumes the amplified portion of write traffic, so at seven times amplification it uses six times the ingest rate in background I/O. Adding that to the foreground writes gives total disk demand, and when it approaches the device limit compaction falls behind: SSTable counts grow, read amplification rises and latency degrades even though ingest looks fine.
Compaction load
compaction throughput = ingest x (amplification - 1); total disk = ingest x amplification; utilisation = total / device throughput
Pending compactions accumulate, each read must consult more files, and eventually the engine applies backpressure or stalls writes entirely.
Throttling protects foreground latency in the short term but guarantees a backlog if the amplified rate genuinely exceeds the device. The real fix is faster storage or lower amplification.