some benchmarks.
old lmdb:
goos: linux
goarch: amd64
pkg: github.com/fiatjaf/eventstore/test
cpu: AMD Ryzen 3 3200G with Radeon Vega Graphics
BenchmarkDBs/lmdb/filter/q-0-4 1000 25128 ns/op
BenchmarkDBs/lmdb/filter/q-1-4 1000 2082 ns/op
BenchmarkDBs/lmdb/filter/q-2-4 1000 3826 ns/op
BenchmarkDBs/lmdb/filter/q-3-4 1000 1017 ns/op
BenchmarkDBs/lmdb/filter/q-4-4 1000 1632 ns/op
BenchmarkDBs/lmdb/filter/q-5-4 1000 2421 ns/op
BenchmarkDBs/lmdb/filter/q-6-4 1000 2091 ns/op
BenchmarkDBs/lmdb/filter/q-7-4 1000 1941 ns/op
BenchmarkDBs/lmdb/filter/q-8-4 1000 1365 ns/op
BenchmarkDBs/lmdb/filter/q-9-4 1000 1173 ns/op
BenchmarkDBs/lmdb/filter/q-10-4 1000 2298 ns/op
BenchmarkDBs/lmdb/filter/q-11-4 1000 1473 ns/op
BenchmarkDBs/lmdb/insert-4 1000 99234 ns/op
new lmdb:
goos: linux
goarch: amd64
pkg: github.com/fiatjaf/eventstore/test
cpu: AMD Ryzen 3 3200G with Radeon Vega Graphics
BenchmarkDBs/lmdb/filter/q-0-4 1000 22811 ns/op
BenchmarkDBs/lmdb/filter/q-1-4 1000 2202 ns/op
BenchmarkDBs/lmdb/filter/q-2-4 1000 3115 ns/op
BenchmarkDBs/lmdb/filter/q-3-4 1000 1285 ns/op
BenchmarkDBs/lmdb/filter/q-4-4 1000 1528 ns/op
BenchmarkDBs/lmdb/filter/q-5-4 1000 2352 ns/op
BenchmarkDBs/lmdb/filter/q-6-4 1000 2824 ns/op
BenchmarkDBs/lmdb/filter/q-7-4 1000 3080 ns/op
BenchmarkDBs/lmdb/filter/q-8-4 1000 1416 ns/op
BenchmarkDBs/lmdb/filter/q-9-4 1000 1383 ns/op
BenchmarkDBs/lmdb/filter/q-10-4 1000 1455 ns/op
BenchmarkDBs/lmdb/filter/q-11-4 1000 1286 ns/op
BenchmarkDBs/lmdb/insert-4 1000 88692 ns/op
so it's not super better but at least it's not worse, and I think in real world it will be better.
badger:
goos: linux
goarch: amd64
pkg: github.com/fiatjaf/eventstore/test
cpu: AMD Ryzen 3 3200G with Radeon Vega Graphics
BenchmarkDBs/badger/filter/q-0-4 1000 1406 ns/op
BenchmarkDBs/badger/filter/q-1-4 1000 1374 ns/op
BenchmarkDBs/badger/filter/q-2-4 1000 3300 ns/op
BenchmarkDBs/badger/filter/q-3-4 1000 848.3 ns/op
BenchmarkDBs/badger/filter/q-4-4 1000 1447 ns/op
BenchmarkDBs/badger/filter/q-5-4 1000 1170 ns/op
BenchmarkDBs/badger/filter/q-6-4 1000 1247 ns/op
BenchmarkDBs/badger/filter/q-7-4 1000 1280 ns/op
BenchmarkDBs/badger/filter/q-8-4 1000 1091 ns/op
BenchmarkDBs/badger/filter/q-9-4 1000 1225 ns/op
BenchmarkDBs/badger/filter/q-10-4 1000 906.7 ns/op
BenchmarkDBs/badger/filter/q-11-4 1000 892.8 ns/op
BenchmarkDBs/badger/insert-4 1000 153319 ns/op
it looks better than LMDB, but I think it's worse in the real world because it uses too much memory and have
all the ristretto cache overhead that looks like it goes not very well with the golang gc, who knows.
sqlite:
BenchmarkDBs/sqlite/filter/q-0-4 50 3910252 ns/op
BenchmarkDBs/sqlite/filter/q-1-4 50 3352545 ns/op
BenchmarkDBs/sqlite/filter/q-2-4 50 12234810 ns/op
BenchmarkDBs/sqlite/filter/q-3-4 50 441476 ns/op
BenchmarkDBs/sqlite/filter/q-4-4 50 179682 ns/op
BenchmarkDBs/sqlite/filter/q-5-4 50 3773350 ns/op
BenchmarkDBs/sqlite/filter/q-6-4 50 3614942 ns/op
BenchmarkDBs/sqlite/filter/q-7-4 50 4040850 ns/op
BenchmarkDBs/sqlite/filter/q-8-4 50 4702405 ns/op
BenchmarkDBs/sqlite/filter/q-9-4 50 5153673 ns/op
BenchmarkDBs/sqlite/filter/q-10-4 50 5840877 ns/op
BenchmarkDBs/sqlite/filter/q-11-4 50 33000137 ns/op
(insert doesn't complete)
by the way these benchmarks are all wrong because they keep halting the computer for some reason. these DBs may not be well-suited for the incessant calling that the go benchmark framework does, I don't know. arbitrarily reducing the number of iterations almost made them work, but tweaking that value yields very different results so everything is useless.
