Hi
Please help. I have got a postgres 7.3.4 database running on RedHat ES 3, with 8GB of physical memory in it. The machine is shared with my application which is pretty intensive in doing selects and updates against the database, but there are usually no more than 10 connections to the database at any time.
Despite having 8GB of RAM on the machine, the machine is frequently running out of physical memory and swapping which is hurting performance. Have read around on various of the message boards, and I suspect that the SHARED_BUFFERS setting on this server is set way to high, and that this in fact may be hurting performance. My current configuration settings are as follows:
shared_buffers = 393216 # min max_connections*2 or 16, 8KB each
max_fsm_relations = 10000 # min 10, fsm is free space map, ~40 bytes
max_fsm_pages = 160001 # min 1000, fsm is free space map, ~6
bytes
sort_mem = 409600 # min 64, size in KB
vacuum_mem = 81920 # min 1024, size in KB
From what Ive read, Ive not seen anyone recommend a SHARED_BUFFERS setting higher than 50,000. Is a setting of 393216 going to cause significant problems, or does this sound about right on an 8GB system, bearing in mind that Id like to reserve at least a couple of GB for my application.
Also if you have any recommendations regarding effective_cache_size Id be interested as reading around this sounds important as well
Thanks
Mark