I believe mongo is supposed to cache rows in memory and simply return results from memory when the working set size is small. I wrote a simple Java client that creates a simple collection with an indexed primary key and a single string field and inserts just one record into the collection, and then repeatedly queries for that record's field using findOne.

The maximum throughput I get with many threads for the above workload is just ~15K/s on a 2-core machine. I clearly can write an in-memory hashmap based cache that can serve reads at nearly ~2 million/s by returning non-dirty entries directly from memory. How can I make mongo do the same? Isn't it automatically supposed to do this optimization? Or is it?

I tried using both mmap and wiredtiger but the read throughput hardly changes. All I need is a big map that automatically pages to disk in the background. Can mongo do this?

PS: Apologies if this is not the right forum for such questions. I tried stackoverflow and dba.stackexchange but didn't get any answers.