On mongo 3.4.9 we are seeing poor performance when the cache hits 95% utilization. I've attached a snippet of mongostat output from a time when this was happening. When this slowdown happens there is almost no disk activity.

Some additional information about our environment:

• The server is running on an r4.8xlarge instance (244GB RAM / 32 cores)
• We are using gzip compression
• We have many concurrent writers and when this slow down happens we max out all of the write tickets. (Is it possible that eviction is having trouble making progress with that many application threads?)
• This collection has 12 indexes.

Are there any configuration settings we should try changing to work around this?

I can upload the diagnostic.data to the private portal if that would help.