Re: oracle parallel server on Pyramid nile 150 & DLM IMPORTANT !!!

That abend looks familiar. The UNIX lock manager processes that the vendor (pyramid in this case) provide to support Oracle Parallel Server OPS require a fair amount of memory per lock. If UNIX does not have enough resources to support what Oracle wants UNIX and the Oracle rdbms do not communicate very well and a system crash ensues. You may want to reduce the number of locks that you are asking for until you have calculated how much memory your parameters require.

With OPS it is very important that you distribute your data such that updates and reader processes come through the same instance as much as possible. We have found that it is possible to separate tables/indexes into tablespaces that are normally attached from one instance by routing user logins to the proper instance behind the scenes. The conflicting processes that can not be eliminated then can be handled by distributing the Oracle OPS locks by file such that the multiple instance accessed tables are supported more locks than files with little conflict. You can use v$ping to help you determine how to distribute the locks best for you environment.

I hope this helps some. We run OPS on Pyramid and we had several unexplained Oracle crashes until we bumped up some supporting UNIX kernal parameters and allocated our locks via files. With a large database you can not even allocate enough locks to support 1 OPS lock per megabyte of data unless you have several gigabytes of UNIX Memory. Received on Tue Sep 17 1996 - 00:00:00 CEST