zilliztech / VectorDBBench

A Benchmark Tool for VectorDB
MIT License
458 stars 108 forks source link

pgvector with postgresql / alloyDB and AlloyDB OMNI #273

Open pauldintel opened 4 months ago

pauldintel commented 4 months ago

With significant performance boost offered by alloyDB postgres Pgvector solution and Zilliz showing open source pgvector worst performance I would like to see if team can publish and measure the benchmark with alloyDB and show competitive results please. thanks

alwayslove2013 commented 4 months ago

Feel free to contribute the AlloyDB client for test and share the performance with us~

pauldintel commented 4 months ago

Feel free to contribute the AlloyDB client for test and share the performance with us~

Sorry if the ask is misunderstood. I am not asking about contribution . Since Zilliz publish the benchmark using VectorDB and pgvector is the last item in the list from performance standpoint . The question is, if you guys may be missing the AlloyDB pgvector part of the story (which Google published as 4x faster on OLTP and 100x faster on analytics) . Does Zilliz having any future aspirations to add alloydb and alloydb omni in zilliz cloud benchmarking results ???

image

alwayslove2013 commented 4 months ago

@pauldintel Currently, we lack sufficient resources to undertake all requested features and enhancements. Therefore, we warmly invite other community members to contribute code or share their results. We truly appreciate your understanding and patience.

xiaofan-luan commented 4 months ago

@alwayslove2013 is there a pgvector support? if there is , then anyone could simply run the test by themselves and we can add the result to leaderboard

alwayslove2013 commented 4 months ago

@alwayslove2013 Yes, vdbbench already supports pgvector and pgvectorrs, and we've posted test results of pgvector as well.

alwayslove2013 commented 4 months ago

In Milestone 1.0.0, we will support testing AlloyDB.

pauldintel commented 2 months ago

And perhaps JVector from Cassandra / Datastax too ?

xiaofan-luan commented 2 months ago

And perhaps JVector from Cassandra / Datastax too ?

We are looking for some volunteers.