Message-ID: <4202AE60.6000706@coretech.co.nz>
Date: Fri, 04 Feb 2005 12:06:08 +1300
From: Mark Kirkwood <markir@coretech.co.nz>
User-Agent: Mozilla/5.0 (X11; U; FreeBSD i386; en-US; rv:1.7.2) Gecko/20041016
MIME-Version: 1.0
To: pgsql-docs@postgresql.org
Subject: Instructions for Linux ipc config
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit

Glancing over these as I was doing the FreeBSD patch, I noticed some
scope for modernization there plus a possible error to correct:

i) we could standardize the change method to use sysctl instead of
echo'ing things into /proc/kernel/* structures (unless there is some
reason for this to be preferable?)

ii) we are recommending setting kernel.shmall equal to kernel.shmmax. I
think this is incorrect or strange anyway :

examining 2.6.10 src I see

./include/linux/shm.h:16: #define SHMALL (SHMMAX/PAGE_SIZE*(SHMMNI/16))
/* max shm system wide (pages) */

i.e. kernel.shmall really is in pages. So our suggested settings of:

kernel.shmall = 134217728
kernel.shmmax = 134217728


allow PAGE_SIZE  128M segments to be created... thats 4096 of 'em on
most common HW. This is probably more than most systems would need or
want (and does setting this value too high have a performance impact?).

One option is to follow what we have said for the BSD's i.e:

kernel.shmall = 32768

or leave shmall at its default (commonly 2097152).

What do people think?


regards

Mark