iiinova — where your digital self keeps evolving

Replies

kero @kerollmops.com · 07/02/2025

The Bustle benchmarks seem to run in parallel, but it's not clear enough in the article. I guessed that, thanks to the Bustle description. Are you running them in parallel, and on what kind of machine/architecture? github.com/jonhoo/bustle 2/4

github.com

GitHub - jonhoo/bustle: A benchmarking harness for concurrent key-value collections

A benchmarking harness for concurrent key-value collections - jonhoo/bustle

200

kero @kerollmops.com · 07/02/2025

In the article, you explain that you need to generate sequential internal ID keys (Meilisearch also uses u32s internally to identify documents). Is this the same use case? What do you use them for? Do you reuse the IDs when deletions happen? And what is the integer size do you use? 3/4

Since internal IDs are always sequential integers

200

Qdrant @qdrant.bsky.social · 07/02/2025

We use u32 as internal ID, whose sequence is unique to each segment.

100

kero @kerollmops.com · 07/02/2025

I wondered if merging the largest memory chunks in the Mask Layer during allocation and deallocation to balance computation on the writing threads is possible. And what happens if you need to allocate a chunk larger than a region? 4/4

100