Linux will try to use any extra RAM for caching the filesystem, and that's what PostgreSQL would like it to do. When the system runs low on RAM, the kernel has a decision to make. Rather than reducing the size of its buffer cache, the OS might instead swap inactive disk pages out. How often to consider this behavior is controlled by a tuneable named swappiness. You can check the current value on your system (probably 60) by looking at /proc/sys/vm/swappiness and the easiest way to make a permanent adjustment is to add a line to /etc/sysctl.conf like the following:
vm.swappiness=0
A value of 0 prefers shrinking the filesystem cache rather than using swap, which is the recommended behavior for getting predictable database ...
