Re: [performance] fast reads on a busy server

On Wed, Jun 27, 2012 at 12:01 PM, Willy-Bas Loos <willybas(at)gmail(dot)com> wrote:

> I cannot follow that reasoning completely. Who needs OS level file cache
> when postgres' shared_buffers is better? The efficiency should go up again
> after passing 50% of shared buffers, where you would be caching everything
> twice.
> The only problem i see is that work_mem and such will end up in SWAP if
> there isn't enough memory left over to allocate.\

That is, 25% probably works best when there is only one cluster.
I'm just wondering about this particular case:
* more than 1 cluster on the machine, no separate file systems.
* need fast writes on one cluster, so steal some memory to fit the DB in
shared_buffers
* now there is useless data in the OS file-cache

Should i use a larger shared_buffers for the other cluster(s) too, so that
i bypass the inefficient OS file-cache?

Cheers,

WBL
--
"Quality comes from focus and clarity of purpose" -- Mark Shuttleworth