This assume that the oracle db is installed under oracle
projadd -c "Oracle" 'user.oracle'
Applies to:
Oracle Server - Enterprise Edition - Version: 9.2.0.1 to 11.2.0.3 - Release: 9.2 to 11.2Oracle Solaris on SPARC (64-bit)
Oracle Solaris on x86-64 (64-bit)
Solaris Operating System (SPARC 64-bit)
Abstract
This is an MOS article that covers kernel parameter setup for Solaris 10 and Solaris 11.
If a Solaris System Administrator needs help implementing these concepts, they will need to contact Oracle's Solaris support organization.
Kernel setup for Solaris 10 and Solaris 11 using project files
Sparc Solaris version 10 and 11
-------------------------------
-------------------------------
Generally speaking, information on setting kernel parameters in Solaris 10 and Solaris 11 is limited. There is a new project file that is used to set kernel parameters (or more generally - any resource control) for the Solaris 10 and Solaris 11 Operating Systems. The remainder of this white paper will be divided into 3 sections to aid in understanding the configuration of kernel parameters in Solaris 10 and Solaris 11.
Basics
Questions and Answers
Examples
Permanently Setting Kernel Parameters
Temporarily Setting Kernel Parameters
Questions and Answers
Examples
Permanently Setting Kernel Parameters
Temporarily Setting Kernel Parameters
Basics on /etc/projects
===================
1. Specifying a limit in the /etc/project file extends that limit to all processes belonging to the project.
===================
1. Specifying a limit in the /etc/project file extends that limit to all processes belonging to the project.
The corresponding ulimit variables that you are concerned with are:
[task|process].max-cpu-time: Maximum CPU time available to processes in this task.
process.max-address-space: Maximum address space.
process.max-core-size: Maximum core dump size.
process.max-data-size: Maximum heap size.
process.max-file-descriptor: Maximum file descriptor index.
process.max-file-size: Maximum file offset allowed for writes.
process.max-stack-size: Maximum stack memory segment available.2. Now, before I show you an example of an /etc/project file containing IPC settings for both the oracle_oltp and oracle_dss projects, please allow me to emphasize that the "oracle_oltp:100:Oracle OLTP........" line below is just one (1) single line with no linebreak. Likewise, the "oracle_dss:101:Oracle DSS......." line below is just one (1) single line with no linebreak. It is only for readability in this whitepaper document that I break these lines up:
oracle_oltp:100:Oracle OLTP:oracle:: \
project.max-shm-memory=(privileged,51539607552,deny);\
process.max-sem-nsems=(privileged,300,deny)
oracle_dss:101:Oracle DSS:oracle:: \
project.max-shm-memory=(privileged,17179869184,deny)This sets a limit of 48GB per shared memory segment and 300 semaphores for all processes in the
oracle_oltp project, and a 16GB shared memory segment limit for all processes in the oracle_dss project.
However, according to the Solaris manual the process.max-sem-nsems gives the maximum number of
semaphores per semaphore set, not for the entire project.
semaphores per semaphore set, not for the entire project.
The issue is that project.max-shm-memory is not an exact equivalent of shmmax.
shmmax is the max size of an _individual_ segment.
shmmax is the max size of an _individual_ segment.
The project.max-shm-memory limit is the __total__ shared memory size for your project.
-- ie maximum total of all your segments.
Similarly, zone.max-shm-memory is the __total__ shared memory size for your zone (see Note:317257.1 for further information)
-- ie maximum total of all your segments.
Similarly, zone.max-shm-memory is the __total__ shared memory size for your zone (see Note:317257.1 for further information)
The seminfo_semmns parameter which used to control the maximum number
of semaphores systemwide is obsolete and has been removed as a tunable
parameter.
So you may want to decrease this number. If you find you run out of these you can increase later.
3. The recommended method for modifying the /etc/project file is to use the "proj*" commands,
such as projadd(1) for creating a project and projmod(1) for modifying a project.
3. The recommended method for modifying the /etc/project file is to use the "proj*" commands,
such as projadd(1) for creating a project and projmod(1) for modifying a project.
Examples of projadd and projmod
# projadd -c "Oracle" 'user.oracle'
# projmod -s -K "project.max-shm-memory=(privileged,6GB,deny)" 'user.oracle'4. Resource Control assignments made in this way (in the /etc/project file) are permanent, and will survive a system re-boot. This is covered in much more detail in the example section below.
5. There is also an "on-the-fly" way to temporarily set Resource Control assignments using the prctl(1) command. However, unlike the /etc/project file, resource assignments made in this way will NOT survive a system re-boot. Again, this is covered in much more detail in the example section below.
6. Oracle support encourages the use of the "id -p" command. It shows the active and available projects for a user. If the limits (such as kernel parameters) that you establish within a project are not being activated for a particular user, the "id -p" will help you to see that the user is not using that project.
============================================================================================================
============================================================================================================
Questions and Answers
===================
===================
Q1: Why can't we just set them the old way?
A1: Actually, you can. While it is not the preferred method, kernel parameter values specified in /etc/system are still honored, with some caveats:
* Values in /etc/system must be greater than the new default values of Solaris 10 / Solaris 11
* Any kernel parameters tunables that are obsolete in Solaris 10 / Solaris 11 are ignored.
* Values specified in /etc/system are global and affect all processes on the system.
* If you use /etc/system to set IPC rctl values, you must reboot the system for them to take effect.
A1: Actually, you can. While it is not the preferred method, kernel parameter values specified in /etc/system are still honored, with some caveats:
* Values in /etc/system must be greater than the new default values of Solaris 10 / Solaris 11
* Any kernel parameters tunables that are obsolete in Solaris 10 / Solaris 11 are ignored.
* Values specified in /etc/system are global and affect all processes on the system.
* If you use /etc/system to set IPC rctl values, you must reboot the system for them to take effect.
As a Note here you still must set noexec_user_stack = 1 in the /etc/system file.
============================================================================================================
============================================================================================================
Real World Examples - Permanently Setting Kernel Parameters
================================================
After upgrading a system from Solaris 9 to Solaris 10 or Solaris 11, it was decided that the new IPC resource control assignment should be used.
The old /etc/system file contained values that were suitable for the Oracle installation on the system,
but these values were too large for average users. The assumption is that all Oracle processes are run under the 'oracle' user.
The contents of the /etc/system file, as pertaining to IPC:
% /bin/egrep "semsys:|shmsys:|msgsys:" /etc/system
set semsys:seminfo_semmni=100 <<< see A.) below
set semsys:seminfo_semmns=1024
set semsys:seminfo_semmsl=256 <<< see B.) below
set semsys:seminfo_semvmx=32767
set shmsys:shminfo_shmmax=4294967295 <<< see C.) below
set shmsys:shminfo_shmmin=1
set shmsys:shminfo_shmmni=256 <<< see D.) below
set shmsys:shminfo_shmseg=10Immediately, these lines can be ignored as the tunables are removed in Solaris 10 and Solaris 11:
set semsys:seminfo_semmns=1024
set semsys:seminfo_semvmx=32767
set shmsys:shminfo_shmmin=1
set shmsys:shminfo_shmseg=10The remaining lines need to have resource controls set up for them. To do so, the first step is that a project must be created. Since all processes are run under the userid 'oracle', the special "user.oracle" project will be used. For more details on special projects, consult the Solaris OS project(4) man page or the Resource Management Guide.
# projadd -c "Oracle" 'user.oracle'A.) Once the project is created, we will assign resource controls corresponding to the remaining lines of the old /etc/system file that need to be "converted". The first line to consider is:
set semsys:seminfo_semmni=100 Since this "100" value is less than the new Solaris 10 default for project.max-sem-ids of "128", we could either artificially lower the value to 100 with a specific resource control assignment, or we could choose to accept the new Solaris 10 default value by simply ignoring the old /etc/system line. We chose to simply ignore it.
B.) The next line of the old /etc/system file that need to be "converted" to a resource control is:
set semsys:seminfo_semmsl=256Again, this value is less than the new Solaris 10 / Solaris 11 default value for process.max-sem-nsems of 512. In this case, however, we wish to artificially limit Oracle to only 256 semaphores per process. Therefore we "convert" this old /etc/system line to a resource control assignment as follows:
# projmod -s -K "process.max-sem-nsems=(privileged,256,deny)" 'user.oracle'C.) The next line of the old /etc/system file that need to be "converted" to a resource control is:
set shmsys:shminfo_shmmax=4294967295This system has 8GB of memory. Therefore, this value (4GB, in bytes) is larger than new Solaris 10 / Solaris 11 default value for project.max-shm-memory of "1/4 physmem" (which would be 2GB on this system). Therefore another resource control assignment must be created:
# projmod -s -K "project.max-shm-memory=(privileged,4GB,deny)" 'user.oracle'D.) The final line of the old /etc/system file that need to be "converted" to a resource control is:
set shmsys:shminfo_shmmni=256Again, this value is larger than the new default for project.max-shm-ids so another resource control assignment needs to be created:
# projmod -s -K "project.max-shm-ids=(privileged,256,deny)" 'user.oracle'Since there are no more lines from /etc/system that pertain to IPC, we remove the old lines:
# /bin/cp /etc/system /etc/system.solaris9
# /bin/egrep -v "semsys:|shmsys:|msgsys:" /etc/system > /etc/system.solaris10
# /bin/mv /etc/system.solaris10 /etc/systemNow, before I show you our final /etc/project file, please allow me to emphasize that the "user.oracle:100:oracle:::process........" line below is just one (1) single line with no linebreak. It is only for readability in this whitepaper document that I break it up:
# cat /etc/project
system:0::::
user.root:1::::
noproject:2::::
default:3::::
group.staff:10::::***********************
NOTICE this is ONE LINE!
************************
user.oracle:100:oracle:::process.max-sem-nsems=(privileged,256,deny);project.max-shm-ids=(privileged,256,deny);project.max-shm-memory=(privileged,4294967296,deny) ***********************
NOTICE this is ONE LINE!
************************
========================================================================
Real World Examples - Temporarily Setting Kernel Parameters
================================================
Resource controls can also be set "on the fly" using prctl(1). Unlike the /etc/project file, resource assignments made in this way will NOT survive a system re-boot. Additionally, the user must have the correct privileges.
The syntax of prctl(1) can, at first, seem complex. Some common usages are:
# prctl -i process to list all resource controls for process
# prctl -i project to list all resource controls for project
# prctl -n -i process lists only the resource control named
# prctl -n -r -v -i process replaces (-r) the named rctl setting with the value
Unlike the /etc/project file, prctl allows the use of "scale factors" to simplify resource control management.
Values specified with the -v switch can be "human readable" values such as 48GB instead of the 51539607552 bytes required in the project database.
For example, assuming the preceding /etc/project file we can check the values for the Shared Memory setting for the oracle_dss project:
% prctl -n project.max-shm-memory -i project oracle_dss
project: 101: oracle_dss
NAME PRIVILEGE VALUE FLAG ACTION RECIPIENT
project.max-shm-memory
privileged 16.0GB - deny -
system 16.0EB max deny -Should we need to temporarily increase the setting to 24GB:
% prctl -n project.max-shm-memory -r -v 24GB -i project oracle_dss
% prctl -n project.max-shm-memory -i project oracle_dss
project: 101: oracle_dss
NAME PRIVILEGE VALUE FLAG ACTION RECIPIENT
project.max-shm-memory
privileged 24.0GB - deny -
system 16.0EB max denyREFERENCES
==========
==========
http://docs.oracle.com/cd/E19082-01/819-2450/rmctrls-1/index.html
The prctl(1) man page and System Administration Guide:
Solaris Containers-Resource Management and Solaris Zones provide several useful examples as well.
The prctl(1) man page and System Administration Guide:
Solaris Containers-Resource Management and Solaris Zones provide several useful examples as well.
