Re: performance tuning in large function / transaction

On Fri, 14 Dec 2001, Christopher Kings-Lynne wrote:

> Hmmm...
>
> I have a database server for a website for which I am 'dedicating' at least
> 128MB of ram (I could say that it can have 256MB)
>
> I have max_connections 64
> and shared_buffers 256
> and sort_mem 1024
>
> Is that really small?
>
> I have this SHM config:
>
> options         SYSVSHM
> options         SYSVMSG
> options         SYSVSEM
>
> options         SHMMAXPGS=16384         # 64MB shared mem?
> #options        SHMALL=1025             # max kb of shared mem
> options         SHMSEG=256              # 256 shared segs per proc
>
> options         SEMMNI=256              # 256 semaphore identifiers
> options         SEMMNS=512              # 512 semaphores in the system
> options         SEMMNU=256              # 256 undo structures in system
> options         SEMMAP=256              # 256 entries in semaphore map
>
> How do you calculate the shared memory required by postgres given the
> shared_buffers value???

IIRC each buffer is 8k, so 256 shared buffers is still pretty small.

sort_mem is wierd due to the way it's used (I think that's per sort, so
maybe you could get more than that per backend if a query had multiple
sort steps), but I think that's backend local memory not shared.