Closed
Description
@LongbinChen pointed out that LMDB would be a good way to support a series of things.
http://symas.com/mdb/microbench/
If we swap out the storage to rely in LMDB we would get a lot of features for free
- Ability to use any key, not just integers
- Out-of-core data (dataset doesn't fit in RAM)
- Decouple index structure from nodes (could potentially set
_K
independently and optimize for it). - Incremental add_item (and deletes) would be much easier to implement
I'm not sure how tricky this would be, but probably not more than 5-10h. Mentioning it in case someone wants to give it a shot :)
Metadata
Metadata
Assignees
Labels
No labels