On Fri, Nov 18, 2011 at 10:41 AM, Tomas Vondra <tv@fuzzy.cz> wrote:
>
> That has nothing to do with the inserts, it means the number of connection
> requests exceeds the max_connections. You've set it to 350, and that seems
> too high - the processes are going to struggle for resources (CPU, I/O and
> memory) and the scheduling gets expensive too.
>
> A good starting point is usually 2*(number of cores + number of drives)
> which is 16 or 24 (not sure what a "dual server" is - probably dual CPU).
> You may increase that if the database more or less fits into memory (so
> less I/O is needed).
Ok, there's just too much conflicting info on the web.
If I reduce the max_connections to 16, how does this reflect on the
Apache MaxClients? There's a school of thought that recommends that
MaxClients in Apache should be the same as max_connection in PGSQL.
But 16 for MaxClients with a prefork MPM would be disastrous. No?
Anyway, even if I do try 16 as the number, what about these settings:
work_mem
shared_buffers
effective_cache_size
With nginx and apache, and some other sundries, I think about 4-5GB is
left for PGSQL. This is 9.0.5. And all other details - Centos 5 on 64
bit, SCSI disks with RAID 10, 3Ware RAID controller...etc.
Any help on settings appreciated.
Thanks!
