Oracle 10g installation on HP UX 11i v2 in Itanium server - Patches & Kernal Parameters needed

[Jahangir Kabir]

I am going to install Oracle 10g Database in a clustered environment. Can you provide me the details for the following?

1. What are the HP UX 11i v2 OS Patches required for installing oracle 10g database R2?

2. What are the ideal kernal parameters and selection methods?

Thanks in advance
Mahbub

[Anshumali]

Hi Jahangir,

1. What are the HP UX 11i v2 OS Patches required for installing oracle 10g database R2?
Ans: there would be a pre-requisite check available from the Oracle Installer where you can easily find the patches needed and the kernel parameter required for the Oracle installation. I hope metalink also have the information for v2 related patches and kernel.
2. What are the ideal kernel parameters and selection methods?
that depends on environment where you intend to plan the setup. Ideal for me may not be suitable for you. Kernel is tuned for performance and its a time based tuning. You can still start with the recommended set of kernel parameter values and tune as per your needs later on by monitoring and deciding on what to change for better performance.
------Sometimes, ideal things are hard to find----

[patrik rybar_1]

hi,

try this oracle note from metalink, there is almost everything include check script
Note:169706.1

or simpliest way is: to run the installer with this flag

-executeSysPrereqs Execute system pre-requisite checks and exit

[Shahul]

Hi,

There is a pre-installer script available from Oracle (I have not run this before, but I have see DBAs produced this to me, which will tell you waht patches are required but not installed and also Kernel parameters require modification.

You may need to note that, Oracle script will pick up sometimes some patches which may not be required in that particular system, for eg, it will alert for patches related to M/C service guard and fiber, etc. If you don't have fiber or M/C service guard, the n you can ignore them.


Kernel parameters, normally picked are these, do not take these values, this is an output from preinstaller script for Oracle 9i.

KSI_ALLOC_MAX set to 65696 is adequate
MAX_THREAD_PROC set to 512 is adequate
MAXSWAPCHUNKS set to 16384 is adequate
MAXUPRC set to 8207 is adequate
MSGMAP set to 30642 is adequate
MSGMNI set to 57484 is adequate
MSGSEG set to 32767 is adequate
MSGTQL set to 82120 is adequate
NCALLOUT set to 8212 is adequate
ALERT- Increase the kernel parameter NCSIZE to at least 35840 from present setting of 22084
NFILE set to 65184 is adequate
NFLOCKS set to 16424 is adequate
ALERT- Increase the kernel parameter NINODE to at least 34816 from present setting of 16964
NKTHREAD set to 30000 is adequate
NPROC set to 8212 is adequate
SEMMAP set to 15322 is adequate
SEMMNI set to 15320 is adequate
SEMMNS set to 30000 is adequate
SEMMNU set to 8208 is adequate
SEMVMX set to 64000 is adequate
SHMMAX set to 1073741824 is adequate
SHMMNI set to 1024 is adequate
SHMSEG set to 512 is adequate
VPS_CEILING set to 1024 is adequate
ALERT- MAXDSIZ set at 2684354560 needs to be increased to at least 1073741824
MAXDSIZ_64BIT set to 3489660928 is adequate
MAXSSIZ set to 391118848 is adequate
MAXSSIZ_64BIT set to 1073741824 is adequate


Hope you got an idea by now.

TIA
Shahul

[Don Morris_1]

Since the context is 11i v2 -- just want to point out that maxswapchunks is not a kernel parameter there anymore (it was obsoleted in v1.6 [11.22]). Don't bother trying to tune it.

I'll also give a caution on that value of maxssiz -- 373Mb of stack is very large for a 32-bit process. Maybe Oracle needs it, maybe not... but it is counter-intuitive with also needing a 1Gb maxdsiz since (assuming default process layout) the stack virtual address space is pre-reserved at process creation time from the same private address quadrant as data grows into. Or in shorter terms -- you can not get more than 1Gb - maxssiz of data for the default process layout. (Moving to text/data in q0 through q1 will get you 2Gb - maxssiz, etc.). Do you really want over 1/3rd of every 32-bit process's private non-text space to be withheld just in case of a large stack?