Hi

We are running RHEL 5.2 with PAE kernel.
Make: HP DL360
Hard Drive Type:Smart Array controller
MemTotal: 16633136 kB
SwapTotal: 33551744 kB
Processor: Intel(R) Xeon(R) CPU 5140 @ 2.33GHz; 2 Dual Core
RedHat Version: Linux SERVER 2.6.18-92.el5PAE #1 SMP i686 i686 i386 GNU/Linux
Application: Oracle 10G

We are facing the oom-killer problem recently on our DB server
(conf:mention above).
Error: Out of memory: Killed process 30700 (oracle).
kernel: oracle invoked oom-killer: gfp_mask=0xd0, order=0, oomkilladj=0

i) the oom-killer process invoked and it kills the process one-by-one it includes oracle, ssh, bash login etc. its because of low memory value

ii) the LOW MEM value gets reduces rapidly when i run Oracle

iii)swap memory is not at all utilised even low memeory is fully drained

work arround we done:
i) set the following parameters
vm.overcommit_memory=2
vm.overcommit_ratio=70
To debug the issue as recommended by some forums but it is not at all solved.
ii)some other suggestion is to set the "vm.lower_zone_protection=100"
But there is no kernel parameter as such in RHEL 5.2. Is there any other alternative param in RHEL5.2 to this param.

If anyone have the idea to solve this issue, pls help us.
Thanks in advance

-Arun

Two questions to start with: did you exactly follow the Oracle installation instructions, when you loaded Oracle (modifying kernel params, etc.), and have you contacted Oracle yet??

Hi
Yes, we configured Oracle installation steps correctly and we contacted Oracle for this issue and they replied its fully an OS issue.
Redhat replied we need to update the 64 bit kernel(now 32 bit) to fix this issue.

Any workarround there?

Thanks
Arun

So ... ???.
PAE was a hack - even Linus hates it; especially for large (> 8 Gig) systems.
The memory regions in 32-bit kernels were for a time and machine architectures long past. Not for big database systems with heaps of memory.

Your answer is to go 64-bit - even if it means new hardware.