Hi @RSommer and welcome in the MongoDB Community 
!
So just to be clear, you are running a 900m*13.6k = 12,24 TB collection on a single standalone node with about 300m/24/60/60 = 3472 writes per seconds in average. Is that correct?
If that’s correct, 12,24 TB is HUGE. Usually we recommend to start Sharding around 2 TB of uncompressed data. We are 6 times over that value.
Can you share more information about the server specs that you are using to support this? I think you just reached the limits of physic and it’s time to shard and distribute the data on multiple nodes.
Also, I guess you have tried and this approach and it’s probably not working for you, but did you try the TTL indexes?
Oh and also, is this a single node Replica Set or just a stand alone node without an oplog?
Cheers,
Maxime.