SHMINFO /usr/lib/dld.sl: Not enough space

[Vic S. Kelan]

Hi!

I am getting the above error while trying to run an application, I believe this is because we just increased SGA on a 32bit oracle recently. We are going to roll back the SGA.

However I ran the SHMINFO tool and have no idea how to interprete the output (attached) How do I know if I have fragmented shared memory pool?

HP-UX 11.11, oracle 8i 32 bit

Thanks!

[Vic S. Kelan]

Sorry my mistake attached is for the system showing the problem.

The other attachment is for a different server

[Steven E. Protter]

Shalom,

May be running into 32 bit limits here.

What is the overall memory on the system? If swap is heavily paging you could have system overload. The cure for that is more RAM, if its not a 32 bit limit 2 GB you are hitting.

Perf data:
http://www.hpux.ws/system.perf.sh

SEP

[Bill Hassell]

The dld.sl message means a shared library (which is part of the one memory map for shared memory in 32bits) can't get enough continguous space to run. SHMINFO shows that the space is fairly fragmented (normal for 32bit apps running at the edge or addressability). It's possible that you need more swap space but most likely you've reached the limits of what you can put in the shared memory map.

Now one common problem is uncontrolled stop and start of applications (like Oracle) using something awful like kill -9. This truly messes up the map so apps can't get the space they need (which must be in contiguous chunks). One fix is to reboot (map starts over) and perhaps the new Oracle SGA will work OK. Another is to implement memory windows which gives selected apps a private map for shared resources. It will require some work to get it all done correctly but such are the growing pains and limits of 32bit programs.

[Vic S. Kelan]

THanks for the info guys!

I guess it must be the shared memory limit I am hitting and I have a better understanding that it is a result of fragmentation of the shared pool....

Here is my swapinfo:

Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 8192 0 8192 0% 0 - 1 /dev/vg00/lvol2
dev 16000 0 16000 0% 0 - 0 /dev/vg00/lvol13
reserve - 5627 -5627
memory 18949 9800 9149 52%
total 43141 15427 27714 36% - 0 -

[Sandman!]

Your swapinfo shows no utilization so the problem has to be with contiguous shared memory allocation. Since increasing the SGA of the Oracle RDBMS would definitely require a bigger and larger shared memory segment then it's possible that a large contiguous chunk of shared memory could not be allocated. Look in the Oracle alert log for startup errors related to shmget.