Write performance vs RocksDB

[vorcigernix]

I picked RocksDB for my side project for following reasons:

Write Performance:
LSM tree structure optimized for write-heavy workloads
Write amplification is minimized compared to B-tree databases
Efficient compaction strategies

Merging performance:
Good for "tree merging" scenarios where identities/records form tree-like structures

Is Limbo something I should put on my radar? My current code is capable of merging ~5k records per second (very simple event-like structures) on my M3 Air, 16GB memory. Thing is that horizontal scaling is kinda problematic with Rocks and while this is clearly a premature optimization to do, I am also continually discovering the options for the day I come to this bridge.

[penberg]

Hey @vorcigernix, we're going to improve write concurrency with MVCC, but are going to stick with B-tree approach. LSM will be better for write-intensive workloads, but 5k records per second sounds like something Limbo should be able to do in the future.

[vorcigernix]

Thank you

[hpvd]

for what is possible today with sqlite when highly optimising:
inserting 500k rows in 0.35s
#2043 (comment)

[hpvd]

There ist now an early technology preview of MVCC implementation in Turso
https://turso.tech/blog/beyond-the-single-writer-limitation-with-tursos-concurrent-writes