Re: Semaphores in dual instance problem

Hi again,

                well, I've finally solved my semaphores problem.... Here's the answers, together with a few things that it wasn't !

1. It was not the shared memory - we had enough shmem specified, and it was being released correctly
2. The instances did *not* start up together, even on boot up !
3. The combinations of the semaphore parameters we were specifying were largely correct, although they were a little over the top in some cases. We now have the following values for parameters

	SEMMNI = 10
	SEMMSL = 25
	SEMMNS = 250

   But, I would stress again that it was not a probelem with    the values of our parameters as specified.

The cause of the problem was that our semaphore parameters were beleived to be taken from the include file

            /usr/include/sys

wheras they were actually being taken from the file

        /usr/kvm/sys/sys

which meant that no matter what values we passed to the semaphore parameters, they were never taken up on the re-build of the kernel !

This was high-lighted by issuing an ipcs -sb command, which showed only one set of semaphores in action and available, even after a re-build using some of the large values for the parameters... Now that we specify them in the correct place, ipcs shows up a healthy collection of 4 sets of semaphores; enough to run both instances of Oracle, and with plenty left over !

My thanks to Roderick Manalac and Joel Garry for their thoughts, which helped me narrow it down, and to Oracle UK support for the same.

                        Simon :-)

Simon Holt 			|	Voice:	+44 (0) 1373 826826 x253
(Systems analyst)		|	Inet:	holt_at_ntrust.org.uk   
	My opinions are not necessarily those of my employer