CPU & RAM Optimisation

[harshakhmk]

Hi, I am looking for ways to optimise CPU RAM Optimisation of Qdrant server, as I have run the server on low end computing machines
Are there any ways to disable certain modules during build to optimise the memory consumption

[generall]

This might be relevant https://qdrant.tech/articles/memory-consumption/

[harshakhmk]

I am interested to know what would be the ideal configurations in config file for an lower end arm architecture machine

[generall]

how much data do you plan to serve there?

[harshakhmk]

I am trying to run the server under 100 MB, and I am experimenting with dataset of 60000 vectors of 785 dimensions

[harshakhmk]

Hi @generall , any comments to optimise resource utilisation for a low end computing machine with a goal to run under 100 MB with a dataset of 20000-60000 with 785 dimensions

[generall]

I run docker run --rm -it --memory=100Mb --network=host qdrant/qdrant:v1.7.4 and it can work pretty much as-is with 60k vectors. You only need to enable on_disk=True for them

[harshakhmk]

Hi @generall , any comments to optimise resource utilisation for a low end computing machine with a goal to run under 100 MB with a dataset of 20000-60000 with 785 dimensions
Could you suggest ideal configuration settings for the above described usecase

To enable on_disk= true, should I add it in client SDK code

[generall]

yes

[harshakhmk]

Hi @generall @dzhao @bazhenov @yarikoptic I was asking for ideal configuration settings for a low end computing machine with a goal to run under 100 MB

[timvisee]

Hi @generall @dzhao @bazhenov @yarikoptic I was asking for ideal configuration settings for a low end computing machine with a goal to run under 100 MB

It'd be best to test this yourself.

But you likely want;

• everything on disk
• only one collection with one shard

Though, just using 100MB of memory will likely make it perform quite badly.