Shared memory.

[Jim Mallett]

I have an L2000 (HPUX 11.00) with 4G of memory and about 3.5G of swap configured.
There are 3 Oracle instances running (1 being Financials).

Any time Oracle is running, the midaemon will not start.
midaemon: Wed Dec 11 10:25:29 2002
mi_create - Shared Memory Database initialization failed; vss = 16993552 bytes
Not enough space.
I get the same type error when I try to run xomni or any other large program on the machine.

I ran swapinfo -tam and it doesn't appear that any of that is even being touched (But I don't know if I'm reading it right.)

I'm going offsite today to do a DR test and during the test I'm going to increase my swapspace to it's 1.5X memory and see if that helps.

I have a 2nd L2000 that has the same setup, but is running 11.11 and has 4G swap (on 1 device) and it runs fine.

I've attached a text file w/ the swapinfo, ipcs, and kmtune info. Just wondering if anybody might be able to see anything screaming "FIX ME"!

[Stefan Farrelly]

It sounds like you dont have enough shared memory configured. What does; kmtune | grep shm
show for sharedmemory size (or use glance/gpm for an overall total) and how much is being used on your server? (ipcs -ma and total the size column).

Looks like you will need to bump up shared memory size on your server, rebuild kernel, and reboot to stop these problems.

[Oleg Zieaev_1]

Hi.

I see your shmmax is small. You might want to increase it.
Are you specifying memory window when starting oracle. You have max_mem_window configured to 3. I think this might impact your total available mem window size while starting 3 oracle instances. If you do not plan to use memory windows, my suggestion would be to switch it to 0.

Hope this helps,
0leg

[John Payne_2]

Yeah, swap space will not likely help you. You 'shouldn't' need to increase your swap to 6GB, especially if you are not using all your 4GB of memory. (Or anywhere close.)

The previous 2 posts will get you there, I think.

Hope it helps

John

[Steven E. Protter]

I agree that the issue in this case is shared memory or message queues.

Make them bigger than you think you need.

If you have a simlar system that works, its parameters are a good starting point.

You need nprocs to be large to run oracle. If you are messing with the kernel anyway, go through the latest metalink document and make sure all parameters mentioned in the doc are in excess of the minimum.

http://metalink.oracle.com

HP-UX 11.11 installs apache and a couple of other nice tools be default. They all use shared memory as well.

Swap however should be increased.

Oracle pushes systems hard and even if its no the cause of this problem, performance issues crop up on systems with less than 16G of RAM that do not follow the good HP rule, swap=twice physical memory.

Steve
Don't forget the points, I collect them.