It's probably helpful if everyone sharing this information can post their measurement process / settings and the results as completely as possible, for comparison and reference.
Apologies. I have only changed one parameter in postgresql.conf for the tests and that was shared_buffers:
shared_buffers = 32GB # min 128k shared_preload_libraries = 'auto_explain' # (change requires restart) vacuum_cost_delay = 5 # 0-100 milliseconds wal_sync_method = open_sync # the default is the first option wal_buffers = -1 # min 32kB, -1 sets based on shared_buffers checkpoint_completion_target = 0.9 # checkpoint target duration, 0.0 - 1.0 checkpoint_warning = 30s # 0 disables default_statistics_target = 100 # range 1-10000 log_line_prefix = '%t ' # special values: log_statement = 'all' # none, ddl, mod, all log_timezone = 'localtime' autovacuum_vacuum_scale_factor = 0.1 # fraction of table size before vacuum autovacuum_vacuum_cost_delay = 5ms # default vacuum cost delay for datestyle = 'iso, dmy' timezone = 'localtime' lc_messages = 'en_ZA.UTF-8' # locale for system error message lc_monetary = 'en_ZA.UTF-8' # locale for monetary formatting lc_numeric = 'en_ZA.UTF-8' # locale for number formatting lc_time = 'en_ZA.UTF-8' # locale for time formatting default_text_search_config = 'pg_catalog.english' auto_explain.log_min_duration = '6s' # Gregory Smith page 180 effective_cache_size = 512GB # pgtune wizard 2014-09-25 work_mem = 4608MB # pgtune wizard 2014-09-25 checkpoint_segments = 16 # pgtune wizard 2014-09-25 max_connections = 80 # pgtune wizard 2014-09-25
# See Gregory Smith: High Performans Postgresql 9.0 pages 81,82 for the next lines vm.swappiness=0 vm.overcommit_memory=2 vm.dirty_ratio = 2 vm.dirty_background_ratio=1 # Maximum shared segment size in bytes kernel.shmmax = 406622322688 # Maximum number of shared memory segments in pages kernel.shmall = 99273028