Two out of three LinkBench indexes are made ~40% smaller with traditional B-Trees after an initial bulk load once this optimization is available: youtu.be/p5RaATILoiE?t=. Were you aware of that? (Only with hacked Postgres that loosens the heuristic.)
Conversation
Replying to
That is good news and I am happy Linkbench is useful. PR are welcome github.com/mdcallag/linkb.
What is the impact on size after a few hours of transactions?
2
Replying to
There are related problems, too (e.g. slightly out of order inserts from IoT devices/sensors). I'm mostly asking about it because it seems like there should be a broad theoretical basis for techniques like this. As far as I can tell, there isn't. Thought you might know something.
1
Replying to
My ability to explain b-tree behavior has faded after a few years of focus on RocksDB
1