ES HEAP // ES infrastructure by 10 billion docs per year

Hi, we use ES 1.4.4. Our Cluster consists of 10 DataNodes within 16 GB Ram total and 10 GB HEAP.
We have 174 indices and 1732 shard (5 primary, 5 replica per index) [we made a new index every month]
Overall we have 818,876,914 docs (replicas are not taken into account) - but we want to expand the system and then save in the future 10 billion docs per year...
We have 3 master nodes, 2 client nodes and 3 proxy clients and use loadbalancing.
Our heap usage increases rapidly and also the retrofitting of Ram || increase the HEAP does not bring long-term improvements more.
Now therefore I ask for advice how to reduce HEAP utilization.

1.) increases the use of the heap linearly with number of docs in the Node / Index / shard?

2.) I have heard the rule: per 100 million docs needed 5-10 GB memory - is this right?

3.) When we generate in the future 10 billion docs per year:
Is this practicable//possible with es [and a not too large number of datanodes] ?

4.) What are the components/functions of the ES-HEAP usage in general? i mean: which functions/services of ES generally uses the HEAP

5.) which are measures/actions to keep the HEAP usage generally lower?

6.) Would you selectively disable / deactivate specific caches to lower the HEAP-usage

7.) Would you hold the caches persistent?

8.) Are there any best-practice settings for the garbage collector?

THX for your help :slight_smile: