I can not start two Oracle 8i instances

[Eugeny_4]

Hello colleagues.
I bump into strange problem. We try to migrate DB from rp7410 server to SuperDome. The both systems have HP-UX 11.11 64bit. Two Oracle 8i 32bit DB work on old rp7410 but I can not start two instances on the new server. Kernel parameters on the SuperDome are equal or greater then the same parameters on the old system. When I try to start second DB on the new unixbox I get :
ORA-27102: out of memory
HP-UX Error: 12: Not enough space
I tried to reduce SGA size but it doesn't help.
In attach You can see kernel parameters of new server.

[Robert-Jan Goossens]

Hi Eugeny,

could you check the ulimit output on both systems for the oracle user?

Regards,
Robert-Jan

[James R. Ferguson]

Hi Eugeny:

Errno 12 can occur if 'maxdsiz' has been met or if you are running out of swapspace. For swap, this would likely mean that you have insufficient space for process swap reservation. Post the output of:

# swapinfo -tam

Regards!

...JRF...

[Victor Fridyev]

Hi Eugeny,

This message means that you don't have enough shared memory configured in kernel. If you may reboot your computer, increase msgmni parameter, otherwise decrease parameter "processes" in $ORACLE_HOME/dbs/init${ORACLE_SID}.ini file.

HTH

[Bill Hassell]

32bit versions of Oracle are extremely limited in memory addressing. There is a single map of 1000 megs (or 1700 megs for SHARE_MAGIC executables) and *ALL* 32 bit programs will try to use the same space. So, if two Oracle instances each use 600 megs for SGA, the second one will fail. And of course, any processes like Oracle and fbackup that use shared memory will leave orphaned segments when these processes are terminated with kill -9. Managing 32bit Oracle instances will need planning which may include implementing HP-UX memory windows.

[Eugeny_4]

Many thanks for pay attention to my problem.
It seems that Bill Hassell is right. But two instances work on the old system. New DBs and old ones have similar initialization parameters. Below I place all parameters mentioned in this thread. db2 is current system (two DBs work). sd1 is new system.

db2 maxdsiz - 53687912
maxdsiz_64bit - 1073741824
msgmni - 50
db2:/ # ulimit
unlimited
db2:/ # swapinfo -tam
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 4096 25 4071 1% 0 - 1 /dev/vg00/lvol2
dev 4096 24 4072 1% 0 - 1 /dev/vg00/lvol9
reserve - 2485 -2485
memory 6309 4962 1347 79%
total 14501 7496 7005 52% - 0 -


sd1 maxdsiz - 1073741824
maxdsiz_64bit - 2147483648
msgmni - 8020
sd1:/\ # ulimit
unlimited
sd1:/\ # swapinfo -tam
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 4096 0 4096 0% 0 - 1 /dev/vg00/lvol2
dev 32768 0 32768 0% 0 - 1 /dev/vg01/lvol1
reserve - 3800 -3800
memory 25442 8071 17371 32%
total 62306 11871 50435 19% - 0 -

[Bill Hassell]

There are likely other processes using shared memory. A reboot will help since the memory maps will be cleared. Shared memory can easily become fragmented with 32bit programs that are improperly terminated. You need to look at the command:

ipcs -bmop

That shows existing segments. You need to compare the shared memory parameters between the two servers:

kmtune | grep shm

(if you are running 11i v2, you'll get a warning about kmtune is now kctune but that's OK)

[Eugeny_4]

On the problems host parameter 'shmmax' is 0x40000000000. On the old 'shmmax' is 0x40000000. Does it cause the problem ? The other 'shm' parameters are equal.

[Sandman!]

>On the problems host parameter 'shmmax' is 0x40000000000. On the old >'shmmax' is 0x40000000. Does it cause the problem ? The other 'shm' >parameters are equal.

Yes that would cause a problem because Oracle will try to allocate a shm segment of that size and error out if it's not available.

[Eugeny_4]

Hmmm... It works. I had to reduce SGA on the both DB but two instance started :)
Thanks a lot.