HP-UX Error: 12: Not Enough Space

[Brian Bientz]

We are currently running Oracle 8.1.7.1 on HP UX 11.00 (both are 64-bit). The instance currently uses about 500 MB for the SGA. We want to bump this to about 1 GB. The system has 2 GB of memory installed. When I try to start the instance with the increased SGA size, I get the following error:

ORA-27102: Out of memory
HP-UX Error: 12: Not enough space

Is this a kernal parameter issue or am I really out of space? Any help would be greatly appreciated.

[Kenneth Platz]

Brian,

Check your "maxdsiz" kernel parameter - this is the maximum permitted size of the data segment of your program.

[Steven Sim Kok Leong · Accepted solution]

Hi,

Run top and swapinfo? It is likely that you have ran out of physical memory. Your SGA size may be too large to fit into whatever memory remains in your physical memory.

For kernel parameters, check the shmmax kernel parameter. Make sure that it is sufficiently large enough to accomodate your Oracle SGA.

From your error messages, it does not seem that you are running low on semaphores or shared memory segments. To be sure however, also perform ipcs -b -m and ipcs -s. Check if there are semaphores or shared memory segments still in used by Oracle ie. not freed up since the last Oracle shutdown.

Hope this helps. Regards.

Steven Sim Kok Leong
Brainbench MVP for Unix Admin
http;//www.brainbench.com

[Brian Bientz]

Seems like there was not enought physical memory. Strange thing is this ... after rebooting, I was able to start up the instance with the larger SGA.

Could there be a memory leak? Or, could the memory have been fragmented?

[Steven Sim Kok Leong]

Hi,

I ever experienced erratic memory usage by Oracle server processes on my system which is running Oracle dedicated server. Instead of each Oracle server process taking up 20 MB memory per established sqlnet connection, it took up 150 MB memory per established sqlnet connection.

Subsequently, we found that the problem was caused by incompatibilities between the Oracle OAS server and the Oracle database server.

First, we noticed that from our swapinfo that our memory was totally consumed while device swap is progressively depleting to the extent whereby the Oracle database server hung and could not be restarted up (similar messages as you had).

From top and glance, I discovered that each Oracle server process was taking up 150 MB memory, unlike the normal 20 MB.

I still find top and glance most helpful, especially glance because they provide ease of access to real-time utilisation of some important kernel parameters (even though you can get this information via an adb script as well) as well as opened files per process (you can also get this via lsof) and other process statistics such as I/O usage per process etc.

I would suggest that you continue using performance monitoring tools especially swapinfo, top and glance to identify the cause of your problem, not to mention checking your logs regularly as well such as Oracle's alert.log and HP-UX's syslog.log.

Hope this helps. Regards.

Steven Sim Kok Leong
Brainbench MVP for Unix Admin
http://www.brainbench.com

[Wodisch]

Hello Brian,
nothing strange about that: the default value of kernel
parameter "dbc_max_pct" is 50 (that is half your RAM).
Stop using it and use "nbuf" and "bufpages" instead,
only assigning as much RAM to the unix buffer cache as
you want (less than 400MB, anyway). Increase the
parameters "shmmax", "semmns" and "maxdsiz", build
the kernel, modify your init*ora parameters, reboot.
HTH,
Wodisch