Adaptive Radix Tree done right
Unlike other implementations in Go/C this version store only different parts of keys. Which leads to dramatically reducing of memory usage in case of storing keys with repeatable fragments
- Optimized memory usage, correct implementation of compact/prefix tree
- Minimized allocations on Set/Get (GC friendly)
- Perfomant
- Store the data in sorted order
- Binary comparator only
WIP
In this example:
tree := art.New()
tree.Set([]byte("http://example.com/tag/10"), []byte("a"))
tree.Set([]byte("http://example.com/tag/20"), []byte("b"))
tree.Set([]byte("http://some.com"), []byte("c"))
t.Log(tree.StringKeys(true))
Tree will looks like:
key:http:// val:
key:example.com/tag/ val:
key:10 val:a
key:20 val:b
key:some.com val:c
$ go test -bench=. -benchmem -count=3 -timeout=1m > x.txt
$ benchstat x.txt
Note: in this benchmark keys are:
- ints - (in bigendian encodings, many common bytes)
- words (more realistic)
name time/op
SetArt-8 115ns ±16%
SetHashMap-8 277ns ±10%
GetArt-8 48.5ns ± 1%
AscendArt-8 182ns ± 8%
DescendArt-8 401ns ± 3%
ScanArt-8 176ns ± 4%
GetHashMap-8 116ns ± 8%
GetWordsArt-8 155ns ± 7%
GetWordsHashMap-8 116ns ± 7%
name alloc/op
SetArt-8 90.0B ± 0%
SetHashMap-8 8.00B ± 0%
GetArt-8 0.00B
AscendArt-8 0.00B
DescendArt-8 263B ± 2%
ScanArt-8 0.00B
GetHashMap-8 0.00B
GetWordsArt-8 0.00B
GetWordsHashMap-8 0.00B
name allocs/op
SetArt-8 1.00 ± 0%
SetHashMap-8 1.00 ± 0%
GetArt-8 0.00
AscendArt-8 0.00
DescendArt-8 0.00
ScanArt-8 0.00
GetHashMap-8 0.00
GetWordsArt-8 0.00
GetWordsHashMap-8 0.00
see tests