Too much I/O generated by postgres stats collector process

Since upgrading PostgreSQL is not an option, I have tried placing the pg_stat_tmp directory on a tmpfs file system, which delivered a significant performance improvement. I am now running this on a few dozen systems for a couple of months without any noticeable drawbacks.

To do this, simply mount pg_stat_tmp with tmpfs in your /etc/fstab file:

# <file system> <mount point>                                <type>  <options>  <dump>  <pass>
tmpfs           /var/lib/postgresql/8.4/main/pg_stat_tmp     tmpfs   defaults,noatime,mode=1777,uid=postgres,gid=postgres,nosuid,nodev 0 0