Qdrant Docker Consuming Full Memory

Hi guys,

Could you guys please help me out?

Scenario: Currently I have 100M of vectors of images. It will increase day by day. We have build a similarity search engine. We have tried with Qdrant Vector Search Engine.

Graviton Instance of AWS: 4 CPUs 32G MEMORY.

Difficulties: When I am initializing qdrant docker it is consuming all of the memory. At a certain point it comes out from the excution. Ia there any way to handle it?Is there any trick with config file or collection creation config.

docker pull qdrant/qdrant docker run -it --memory="26g" --memory-swap="200g" --memory-reservation="22g" -p 6333:6333 -v $(pwd)/../qdrant/data:/qdrant/storage \ -v $(pwd)/data/static/qdrant_config.yaml:/qdrant/config/production.yaml \ qdrant/qdrant:latest.

Hi @Shaibal , I assume you have something like 512d vectors. With this amount of data you would need to keep almost all vectors on disk, which will make search quite slow. I would recommend to run the cluster of several machines for this amount of vectors

Yes its 648d vectors.

How should I build my architecture then? Could you please brief a little bit?

Are you talking about distributed system?

yes

the easiest - to use our cloud. We provide multi-machine configuration out of the box

https://cloud.qdrant.io/

Could you please elaborate it a little? I will discuss with my team.

Could you please let me the know the cost?

depends on your latency requirements.

How much cost it will take consider our vectors will increase day by day.

Lets consider 100M vectors of 648d.

Could you please give an idea?

how fast do you want the search to be? It terms of RPS or ms per request?

I need 15-20RPS.

something like this should work

Ou great! Can I have access of your costing calculator?

yes, here https://cloud.qdrant.io/

Is it available on your qdrant.tech site?

you would need to register

Ou I see.

Thank you @andrey.vasnetsov. It's a nice discussion.