ML Node Java Heap Ram Ratio

Hi,

I want to add a dedicated ML node to my cluster. Unfortunately, I couldn’t find any recommendation what Java Heap to Ram Ratio should be used. I know that on a normal node 50% of the Ram should be used for the Java Heap. Is this number the same for a dedicated ML Node?

Thank you very much.

There isn’t a fixed rule like the 50% Java heap to RAM ratio typically recommended for normal nodes. But we are using the same for ml-node. Example docker compose file

1 Like