hello Jaim,
Regarding cache-line-contention issue , I think you can refer this page
Once I resized my shared_buffer from 15GB to 45GB, AP team reported performance issue after 2 days ,
and we found swap in/out activities increased vastly
At that time, the only thing I can do was returning to the original shared_buffer setting -- 15GB,
and performance recovered to normal.
I consulted EDB for memory issue, because their slides introduced at most 8GB for shared_buffer.
And I got positive response from them at that time.
Now I think it maybe an numa issue.
I will try the same evualtion size of data buffer size as Oracle's SGA in my future projects.
Or anybody can share his/her experiences with large shared_buffer size with postgres db.
Regards,
Steven