Oh yeah, sorry. My box has 1.5 gig ram, but it is an application server
that runs things other than just postgresql. It also runs:
Apache
Real Server
OpenLDAP
Squid
Samba
with all those services fired up and running, as well as postgresql with
256 Megs of shared buffer, I have about 900 megs of cache and 100 megs
free ram. Since a lot of data is flying off the hard drives at any given
time, favoring one service (database) over the others makes little sense
for me, and I've found that there was little or no performance gain from
256 Megs ram over say 128 meg or 64 meg.
We run about 50 databases averaging about 25megs each or so (backed up,
it's about 50 to 75 Megs on the machine's hard drives) so there's no
way for ALL the data to fit into memory.
On Tue, 7 Jan 2003, Fred Moyer wrote:
> To put this usage of shared buffers in perspective would you mind kindly
> let us know your total amount of system ram? Without hearing what
> percentage of memory used as shared buffers (assuming is the primary
> application being using here)
>
> I have always taken the 'more is better' approach with shared buffers but
> would like to know what in terms of percentages other people are using. I
> have been using 50% of system ram (2 out of 4 gigs) for shared buffers
> (and corresponding shmmax values) and it has been working great. I
> haven't tweaked the kernel yet to get more than 2 gigs shmmax so I can't
> speak for a setup using over 50%. I've been using between 256 and 512
> megs sort memory which sounds like a little much from what I'm hearing
> here.
