[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference turris::digital_unix

Title:DIGITAL UNIX(FORMERLY KNOWN AS DEC OSF/1)
Notice:Welcome to the Digital UNIX Conference
Moderator:SMURF::DENHAM
Created:Thu Mar 16 1995
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:10068
Total number of notes:35879

9439.0. "Problems modifying kernel parameters" by OZROCK::MOORED () Wed Apr 09 1997 03:38

I have a program that performs stress testing on my 
application.  The stress test environment is affected
by setting kernel subsystem parameters
proc:
    max-proc-per-user = 1500
    max-threads-per-user = 3000

ipc:
    msg-mni = 1024
    msg-tql = 1050

These parameter values are set by directly editing
/etc/sysconfigtab and rebooting the system.

A kernel rebuild was done to increase maxusers to 512.  To do this I:
cd /sys/conf
vi BORICH  (BORICH is the name of the machine being used)
doconfig -c BORICH 
(did not edit the kernel config file during rebuild)
cp /vmunix /vmunix.sav
mv /sys/BORICH/vmunix /vmunix

The machine was rebooted with shutdown -r now.


After reboot I successfully ran my stress tests. GREAT!

To make the stress test fail I can change almost any of the
values in /etc/sysconfigtab mentioned above to a lower value.
I lowered the value of msg-tql = 800 and rebooted the machine.  The stress
tests failed as expected.

The value of msg-tql = 1050 was set (same as before) and machine rebooted.
The stress test unexpectedly failed.

In an attempt to get the test to pass again the kernel was rebuilt but without
modifying /sys/conf/BORICH.  

The new /sys/BORICH/vmunix was the same size as /vmunix but executing
diff /sys/BORICH/vmunix /vmunix returned a message indicating the binary files differ
(Is that significant, given that /sys/conf/BORICH had not changed ?)

/sys/conf/BORICH was moved to /vmunix and the machine was rebooted.  The stress test passes
again.  If I change a value in /etc/sysconfigtab and then change it back the problem
reoccurs.  Why can't I expect to see my test behave in a pradicable way when I have what should be
a known state in terms of values /etc/sysconfigtab file. ?

Do I need to rebuild the kernel after making changes to /etc/sysconfigtab or is rebooting
sufficient for changes to take effect ?

uname -a returns OSF1 borich.ozy.dec.com V3.2 214 alpha
T.RTitleUserPersonal
Name
DateLines
9439.1More InfoOZROCK::MOOREDWed Apr 09 1997 04:102
After modifying /etc/sysconfigtab and rebooting the new value
for msg-tql was verifed with sysconfig -q ipc.