MongoDB server version: 3.4.10
OS: Debian 4.9.65-3+deb9u1~bpo8+1 (2017-12-23) x86_64 GNU/Linux
RAM: 32GB
CPU: 12 cores
Database size: 61GB (no replication, no sharding)
Collection records: 32 Millions (only one collection in the database)

On server peak hours something hinky happens. Some queries on the database takes from 10s to 50s! When I get htop, all of a sudden can see that mongod has used ONLY 40% of server RAM and about 14GB of memory is free, cpu usage is used about 10% to 20%! When I see those queries I can see that it has used `IXSCAN` as its winning plan and fetched like 4000 records which belongs to the user, which is normal.

Why `mongod` has not used all of the memory and causes disk IO? It has been mentioned in the documentation that mongod will consume all the memory on the server!

I should note that I have benchmarked disk IO using `hdparm` and result was like below:

root@hs1.example.com:~# hdparm -tT /dev/mapper/BDRS--CH01--001--vg-VAR
/dev/mapper/BDRS--CH01--001--vg-VAR:
 Timing cached reads:   17388 MB in  2.00 seconds = 8706.52 MB/sec
 Timing buffered disk reads: 2722 MB in  3.00 seconds = 907.05 MB/sec
 
root@hs1.example.com:~# hdparm -tT /dev/mapper/BDRS--CH01--001--vg-VAR
/dev/mapper/BDRS--CH01--001--vg-VAR:
 Timing cached reads:   17022 MB in  2.00 seconds = 8523.85 MB/sec
 Timing buffered disk reads: 2054 MB in  3.01 seconds = 683.17 MB/sec
 
root@hs1.example.com:~# hdparm -tT /dev/mapper/BDRS--CH01--001--vg-VAR
/dev/mapper/BDRS--CH01--001--vg-VAR:
 Timing cached reads:   17732 MB in  2.00 seconds = 8879.00 MB/sec
 Timing buffered disk reads: 3298 MB in  3.00 seconds = 1099.23 MB/sec

This misbehaviour seems as a bug not normal behaviour.