On Fri, Dec 11, 2009 at 6:40 AM, Thorne, Francis <thornef@cromwell.co.uk> wrote:
> Hi All
>
> I'm looking for some help with regards a postgres 64-bit install I set up
> on a IBM AIX machine.
>
> I used the following install options whien I installed postgres
>
> BINDIR =
> /usr/local/pgsql837-64/bin
> DOCDIR =
> /usr/local/pgsql837-64/doc
> INCLUDEDIR =
> /usr/local/pgsql837-64/include
> PKGINCLUDEDIR =
> /usr/local/pgsql837-64/include
> INCLUDEDIR-SERVER =
> /usr/local/pgsql837-64/include/server
> LIBDIR =
> /usr/local/pgsql837-64/lib
> PKGLIBDIR =
> /usr/local/pgsql837-64/lib
> LOCALEDIR
> =
> MANDIR =
> /usr/local/pgsql837-64/man
> SHAREDIR =
> /usr/local/pgsql837-64/share
> SYSCONFDIR =
> /usr/local/pgsql837-64/etc
> PGXS =
> /usr/local/pgsql837-64/lib/pgxs/src/makefiles/pgxs.mk
> CONFIGURE = '--prefix=/usr/local/pgsql837-64' '--with-pgport=5422'
> '--enable-thr
> ead-safety' '--enable-integer-datetimes' 'CC=gcc -maix64'
> 'LDFLAGS=-Wl,-bbigtoc'
> CC = gcc
> -maix64
> CPPFLAGS
> =
> CFLAGS = -O2 -Wall -Wmissing-prototypes -Wpointer-arith -Winline
> -Wdeclaration-a
> fter-statement -Wendif-labels -fno-strict-aliasing
> -fwrapv
> CFLAGS_SL
> =
> LDFLAGS = -Wl,-bbigtoc
> -Wl,-blibpath:/usr/local/pgsql837-64/lib:/usr/lib:/lib
> LDFLAGS_SL = -Wl,-bnoentry -Wl,-H512
> -Wl,-bM:SRE
> LIBS = -lpgport -lz -lreadline -lld
> -lm
> VERSION = PostgreSQL 8.3.7
>
> And the spec of the AIX machine is a follows.
>
> AIX 5.3
> 2x Power 5 CPU's
> 6GB RAM
>
> If I run postgres with the standard postgresql.conf install then all queries
> run (albiet sometimes slower than i'd expect). However if I start to change
> some of the memory settings for example shared_buffers, temp_buffers,
> work_mem, maintenance_workmem above 128-256MB and run a large query or
> create a large indexes i get the error below
>
> Out of Memory
> Failed on request of size 167873 (this value will change)
>
> Any thoughts or ideas would be greatly appreciated
Don't set work_mem so high? shared_buffers at 256M is no big deal.
work_mem at 256M is per sort. So a dozen users each running queries
with 3 or 4 sorts could chew up 6G pretty fast.
