ScyllaDB
Compaction Strategy
Simulator
Step
Pause
Reset
Compaction strategy
LCS — Leveled
ICS — Incremental
TWCS — Time-Window
Workload
Overwrite-heavy (80/20 hot keys)
Uniform random keys
Time series (new key each write)
⚠
×
min_threshold
4
sstable_size_in_mb
1000
space_amplification_goal
null (off)
1.25
1.5
2.0
i
×
sstable_size_in_mb
160
compaction_window_size
1
compaction_window_unit
24 h
MINUTES
HOURS
DAYS
Speed
1×
Write path
Client
0
writes issued
→
Commitlog
0 MB
append-only, truncated on flush
→
Memtable
0 / 12 MB
in RAM, sorted by key
→
Flush
0
SSTables written
→
Compaction
0
idle
SSTables on disk
newer / smaller → older / larger
selected for compaction
being written
block width ∝ size on disk
Amplification
SSTables
0
On disk
0
MB
Live data
0
MB
unique live rows
Space amp
1.00
×
peak 1.00×
Write amp
1.00
×
bytes written / flushed
Read amp
0
files per lookup
Amplification over time
Space
Write
Read
Strategy comparison
typical at production scale — the tiles above measure
this
run