Hi all,
my appologies in advance as this is not a tru64 unix question,
however we don't have any support for this machine and I haven't
been able to find a email group for unixware.
We have a system running Unixware 7.1.0. On this system we
have a 3rd party software (acu cobol based) which uses "bequeath"
to access an oracle database (oracle v8.05). (ie. the software
doesn't use the oracle listener and hence isn't using tcp/ip).
The problem is :
At numerous times during the day users are unable to get into
the system until someone else logs out. (Error 12223)....and
the users are starting to get a bit annoyed (to say the least)
The 3rd party software support seems to think it is either an oracle
or unix problem.....
>- Originally believed problem to be the number of users logging into the
>database as "gcccuser". Created a new user "gcccmenu" and appropriate cblconfig
>file (/etc/cblconfig.SYS) to try and spread the load. Now whenever a user logs
>into the system they login as "gcccmenu" then when they run a menu option they
>use "gcccuser". Tested until 12223 error re-occurred. When error occurred
>(little difference in number of users) got the database administrator to log
>into unix and access the database using "sqlplus" bypassing Acucobol, obtained
>the same 12223 error.
>
>Rebuilt "runcbl" and generated a smaller executable. This allowed more users
>onto the system however was still stopping after about 87 processes (menu).
>Checked if user could login using "sqlplus" and 12223 error was occurring
>during this process.
>
>- With difference in number of processes and size of executable I now believe
>that the problem may now be "Shared Memory" or Memory related in some way.
>Due to the fact that using standard sqlplus and unix obtains the
>same 12223 error I believe that the problem is Kernel/Oracle related and I tend
>to think that it is more an Oracle problem than Unix related.
Oracle support however, don't believe it is an oracle problem.
I am new at unixware, so am unsure if it is a unix
problem.....the error from oracle points to it running out of
system resources (ie. memory), but looking at the system at the
time of the error.....it definitely isn't running out of memory.
I am not an Oracle DBA, so don't understand oracle too well, but
below are the notes from our DBA who has been working with me
on this problem.
The following is the status with trying to solve the problem of 9D error on
licensing:
1. Truss log created for Unix calls. Sqlnet log created for Oracle calls.
2. When 9D error occurs we can log onto UXORAC01 and fork as many processes as
we want.
It does not appear to be process related.
However we cannot log into Oracle when the 9D error occurs. A resource that
Oracle requires has
been exhausted.
3. Oracle (tar 9399088.700) advised that after looking at the truss log the
error relates to ENOMEM(Error 12)
which is memory related.
There is plenty of memory at the time of the 9D so it is not running out of
memory but rather it is a
parameter related to memory that has reached its maximum.
4. The doco on ENOMEM points to parameters SDATLIM, HDATLIM,SVMMLIM,HVMMLIM.
Theses
parameters have been set to the maximum.
However SSTKLIM which relates to SVMMLIM was not at its maximum so we
increased this to the maximum.
This made matters worse so we reverted back to the original
5. We have increased SHMMNI from 800 to maximum of 1000....with no change
Attachements...............
[a] Attached is the truss of the problem that we are having. The SIGHUP is
when the user
end tasked the screen. Before that it just hangs.
[b] Attached is the client sqlnet client trace of the TNS-12223 error.
The following is the output from the truss
*** Expected fork SYSEXIT
*** cannot trace across exec() of /usr/users/app/oracle/product/8.0.
5/bin/oracle ***
*** Expected fork SYSEXIT
*** Expected fork SYSEXIT
*** Expected fork SYSEXIT
*** Expected fork SYSEXIT
The truss command was :
truss -faic -sall -o truss.out acumenu
Oh, and we are not running out of licenses......checked that :)
And the total number of user processes on the system
has varied from ~170 to 300+ at the times that the error occurs.
Any ideas or suggestions would be greatly appreciated.
thanks
Ruth :)
Ruth Reeve
Unix System Administrator
Gold Coast City Council
Phone: (07) 55816955
Fax: (07) 55740700
Mobile: 0414 180908 or 4908
E-mail: rreeve_at_goldcoast.qld.gov.au
**********************************************************************
This e-mail and its contents is confidential to Gold Coast City Council
and un-authorised use is strictly prohibited.
**********************************************************************
Received on Thu Aug 24 2000 - 04:10:29 NZST