I have today found we have installed 2 2-node clusters in our enterprise environment, where each server contains 2 Terabytes of physical memory. (YES, TWO TERABYTES).
I have always tried to go by the rules for the following parameter settings,
page size is 2048Kb
4K page size 4096Kb
for /etc/sysctl.conf
kernel.shmmax = Range 50% physical memory +1 to 99% physical memory +1 (always stick to 50%)
kernel.shmmall(#pages) = physical memory / 2 / 4K page size
vm.nr_hugepages(# hugepages) = physical memory / 2 / 2M huge page size
for /etc/security/limits.conf
oracle soft nofile 65,536
oracle hard nofile 65,536
oracle soft nproc 16,384
oracle hard nproc 16,384
oracle soft stack 10,240
oracle hard sttack 32,768
oracle soft memlock = physical memory (Kb) * 90%
oracle hard memlock = physical memory (Kb) * 90%
Our systems are dedicated to Oracle software only; no application software running on the servers. The os is OEL 6.[56] with UEK kernel enabled.
We have historically setup systems with 128Gb, 256Gb, and 512Gb, and use the above parameter settings as our standard. Now we have a couple servers running 2Tb physical memory.
Do the same rules apply regardless of the amount of physical memory or do they change? Have I got any of the above formulas wrong?
David Riskind