Re: performance problems.

On Wed, 30 Aug 2006, Jim C. Nasby wrote:

> On Wed, Aug 30, 2006 at 10:10:28AM -0400, Vivek Khera wrote:
> > effective_cache_size = 27462 # `sysctl -n
> > vfs.hibufspace` / 8192 (BLKSZ)
> > random_page_cost = 2
>
> You misunderstand how effective_cache_size is used. It's the *only*
> memory factor that plays a role in cost estimator functions. This means
> it should include the memory set aside for caching in shared_buffers.
>
> Also, hibufspace is only talking about filesystem buffers in FreeBSD,
> which AFAIK has nothing to do with total memory available for caching,
> since VM pages are also used to cache data.

I believe it's not talking about quantities of buffers at all, but about
kernel virtual address space. It's something like the amount of kernel
virtual address space available for mapping buffer-cache pages in to
kernel memory. It certainly won't tell you (or even approximate) how much
PostgreSQL data is being cached by the OS. Cached PostgreSQL data will
appear in the active, inactive and cached values - and (AFAIK) there isn't
any distinction between file-backed pages and swap-backed pages amongst
those.