On Wed, Jan 27, 2010 at 04:38:08PM -0700, Jeff Ross wrote: > Tobias Ulmer wrote: > >On Wed, Jan 27, 2010 at 02:13:45PM -0700, Jeff Ross wrote: > >>I have searched (and searched) so I wonder if I'm running into the > >>i386 1GB limit I see referenced, as in the thread today about fsck > >>on larger partitions. > > > >Yes you do. Also, kernel memory is limited, insane shm value will > >probably (havn't looked at the code) have bad effects. > > > > Thanks! > > By what definition of insane? I'd like to be able to say to the > PostgreSQL folks that their sizing model doesn't work on OpenBSD > because of x but this is the correct way to size on OpenBSD. > > Also, I just saw a message from in the same thread referenced above > that amd64 is also limited to, if I'm reading between the lines > correctly, 3GB physical memory. > > The postgresql people suggest it is a ulimit problem but I have > followed the README.OpenBSD there as well and put the _postgresql > user in its own login class with increased openfiles-cur limits. > Unfortunately that has not helped. > > Jeff
Let's make this simple, on i386 you have 1G per process. Adding all the numbers up, you have to stay below this limit. If that's not enough, get amd64 and replace 1G with 8G. Keep in mind that virtual memory needs to be backed by physical mem and/or swap if you want to use it. In other words: available = MIN(datasize, physical memory);
