yurib wrote:There is a bug in Red Hat Enterprise Linux and Oracle Linux 6 that means changes to /etc/fstab for /dev/shm are not seen during boot. If you want to change the size of /dev/shm, you need to manually remount it in /etc/rc.local. For example:
Can I change the */etc/fstab* entry for /dev/shm to manually increase the size to take up ~80-90% of the server physical RAM ? Is it a good idea?
However, you have 64GB of RAM, so you should look at disabling AMM and using HugePages instead. It's way more efficient for any server with >4-8GB of RAM. HugePages don't use /dev/shm.
# mount -o remount,size=50GB /dev/shm
yurib wrote:Our Director of Sustaining Engineering lists not configuring HugePages as the #1 cause of Database issues: http://event.on24.com/r.htm?e=452104&s=1&k=9292BFF32FC2A347E4771667EC0502D5&partnerref=OracleLinux_Linux_US
I have not seen any definitive performance studies showing me how much faster the system will perform with huge-pages so we are leaning towards AMM and setting some minimum SGA and PGA values as a baseline.
yurib wrote:It's not a kernel bug, so it's present in all versions of OL6.
is this bug present in 6.2 with UEK R2 also ?
to make this /dev/shm change using umount - do I need to first shut down Oracle 11g R2 and then shut down Oracle Grid/ASM and then use umount?I've edited my post - you need to use the mount command, not umount. My apologies for that. You don't need to shut anything down to change the size of /dev/shm.
what is some other part of OS/app is using the /dev/shm? How do I find out and shut them down so I can re-configure /dev/shm?This is not required.
It simply removes the -f parameter when mounting /dev/shm in rc.sysinit. Then after a reboot /dev/shm should be sized according to /etc/fstab and you don't have to remount it.
sudo cp /etc/rc.d/rc.sysinit /etc/rc.d/rc.sysinit.bck sudo sed -i 's/-f \/dev\/shm/\/dev\/shm/g' /etc/rc.d/rc.sysinit