Oracle RAC - sysctl.conf, Oracle block

category: solutionz · date: 2016-12-31 · updated: 2026-10-02 · author: LALA

Oracle RAC Solution · Config document · referenced from Operating system preparation

notevm.hugetlb_shm_group is set twice in this block. Only the second value is in effect; do not copy both lines.

The kernel parameters added to /etc/sysctl.conf for Oracle Grid Infrastructure and Oracle Database 11.2: the values of the installation guide, shared memory sized for the machine, the group allowed to use huge pages, and two network settings for the interconnect interface. The notes show only this added block, not the rest of the file.

Item	Value
Path	`/etc/sysctl.conf`
On	`oradb01` and `oradb02`
Shown here	the block added to the existing file; the notes show `...` above it
Activated with	`sysctl -p`, as `root`
Operating system	SUSE Linux Enterprise Server 11 SP3, kernel 3.0.76-0.11-default

The file

ini

# ... (the existing file continues above; the notes do not show it)
# Kernel parameters for running Oracle Database
fs.aio-max-nr = 1048576
fs.file-max = 6815744
kernel.shmall = 2097152
kernel.shmmax = 4124030976
kernel.shmmni = 4096
kernel.sem = 250 32000 100 128
net.ipv4.ip_local_port_range = 9000 65500
net.core.rmem_default = 262144
net.core.rmem_max = 4194304
net.core.wmem_default = 262144
net.core.wmem_max = 1048576

# Let the local group oinstall (30001) create shared memory segments (SUSE Linux only.)
# Installation.
vm.hugetlb_shm_group=30001

# Let the local group dba (30004) create shared memory segments (SUSE Linux only.)
# Operation.
vm.hugetlb_shm_group=30004

# Here the adapter "eth3" (eth3:INT) is used as the Oracle interconnect
net.ipv4.conf.eth3.rp_filter = 0
net.ipv4.icmp_echo_ignore_broadcasts = 0

Line	Meaning	Where the value comes from
`fs.aio-max-nr`	system-wide limit of asynchronous I/O requests	the guide's minimum
`fs.file-max`	system-wide limit of open file handles	the guide's minimum
`kernel.shmall`	total shared memory, in pages	the guide's minimum; the installer rejected the 1006843 that `shmsetup.sh` had calculated
`kernel.shmmax`	largest single shared memory segment, in bytes	the value the installer asked for; `shmsetup.sh` had calculated 4124028928, the guide's minimum is 536870912
`kernel.shmmni`	number of shared memory segments	the guide's minimum
`kernel.sem`	semaphore limits: per set, system-wide, operations per call, number of sets	the guide's minimum
`net.ipv4.ip_local_port_range`	ephemeral port range	the guide's value
`net.core.rmem_default`, `rmem_max`, `wmem_default`, `wmem_max`	default and maximum socket buffer sizes	the guide's minimums
`vm.hugetlb_shm_group`	the one group whose members may create shared memory segments backed by huge pages	30001 is `oinstall`, 30004 is `dba`
`net.ipv4.conf.eth3.rp_filter`	reverse path filtering on `eth3`, switched off	added when the interconnect became the alias `eth3:INT`
`net.ipv4.icmp_echo_ignore_broadcasts`	0 lets the node answer pings sent to a broadcast address	same comment in the file; the notes give no further reason

vm.hugetlb_shm_group takes one group ID. sysctl -p applies the lines in the order of the file, so the first assignment (30001, "installation") is overwritten at once by the second (30004, "operation"). The comments suggest the value was meant to be switched after the installation; the notes show both lines active.

The last two lines mention eth3:INT, which exists only since the interconnect was moved from eth2 to an alias on the management interface. They were therefore added later than the rest of the block, during the events of Grid patches and the multicast problem.

Checked against Oracle AI Database 26ai

As built	Today
`fs.aio-max-nr`, `fs.file-max`, `kernel.shmmni`, `kernel.sem`, the four `net.core` values	Same minimums in the 26ai table
`kernel.shmall = 2097152`	No fixed number any more: "greater than or equal to the value of shmmax, in pages". 2097152 satisfies that for this `shmmax`
`kernel.shmmax = 4124030976`	"Half the size of physical memory in bytes", which is what the installer asked for here
`net.ipv4.ip_local_port_range = 9000 65500`	Minimum 9000, maximum 65535
not set	`kernel.panic_on_oops = 1` is in the table since 12c, `kernel.panic = 10` is new in 26ai
`vm.hugetlb_shm_group` twice	Oracle, in 11.2 and in 26ai: enter the GID of the `oinstall` group, and "only one group can be defined". The second line wins, so the effective group was `dba`, of which `grid` is not a member. Without `vm.nr_hugepages` it had no practical effect. The 26ai procedure puts the setting into `/etc/sysctl.d/99-hugetlb-shm-group.conf`
`net.ipv4.conf.eth3.rp_filter = 0`	Oracle recommends 0 or 2 for private interconnects and requires it only with several of them. The kernel uses the higher of the `all` and the per-interface value, so 0 on one interface is ineffective if `all` is 1; 2 is the safe value. SUSE's default is already 2
`net.ipv4.icmp_echo_ignore_broadcasts = 0`	Not an Oracle requirement in the 11.2, 19c or 26ai guides. It only makes the host answer broadcast pings; it does not make a network carry broadcast
the block in `/etc/sysctl.conf`, `sysctl -p`	Still works. A file in `/etc/sysctl.d/` and `sysctl --system` is the preferred form
raising `shmmax` and `shmall`	Since Linux 3.16 both default to nearly the maximum, so on SLES 15 and later Oracle's values lower the limits. The installer still checks them

The eleven Oracle values can still be used as they are. The last three settings of the block should not be copied: one of the two vm.hugetlb_shm_group lines is dead, and the two network lines belong to the diagnosis of a network that passed no broadcast, not to a normal installation.