Conference ulysse::rdb_vms_competition

>    I know that ORACLE performance is severly hampered in a CI-Cluster
>    because of a Master-Slave relationship. But how does ORACLE performance
>    degradation in a LAVC compare with Rdb in the same situation?
         
Paul,

I don't know what you mean by a 'master-slave' relationship within Oracle.

Oracle relies upon the extensive use of shared, global memory that they
call the SGA (system global area).  The preferred method is to restrict
access to a database to a single node in the cluster and thus to
avoid using the VMS lock manager and internal lock managers.

As other nodes are given access to the Oracle database, whether or not
they take advantage of that access, Oracle does more locking.

In addition, because Oracle tries to use a fast commit, deferred write approach,
as other nodes access the database, ASTs (or some such construct) are used
to signal the release of data from SGAs.

Between more extensive use of locking and ASTs, Oracle performance rapidly
deteriorates on a cluster.  

My understanding is that a LAVC is just like a CI-Cluster except that 
communication is done of the slower ethernet rather than the CI bus.
Therefore, the performance problems will be increased for Oracle is such
a situation.

While Rdb is not a screamer in heavy update applications using a cluster,
it is designed around a distributed processing architecture.  That is, it
has been designed to use the VMS lock manager and ASTs whereas Oracle is
not.  Thus, in cluster situations, Rdb is a much better performer.

There are a number of ways to balance Rdb performance in a cluster.

	- Controlling the number of buffers for each Rdb image can
	  reduce the number of ASTs

	- using read-only transactions and snapshots can reduce locking
	  
	- using the adjustable lock granularity (default) locking scheme
	  minimizes the amount of locks.

	- reducing the number of active database attaches (ACMS) can
	  reduce the number of locks

	- use RMU to monitor lock activity and restructure databases and
	  applications.


I hope this is enough info.  Call if you have more questions.

--david

    I guess I meant to say ORACLE used centralized .vs. distributed     
    management of the database. But you have answered my question. ORACLE
    uses the same centralized scenario in a LAVC. Thus degraded perforance.    
                                                                           
    If the node that is managing the ORACLE database goes down, the        
    remaining users lose access to the database, (no failover), Correct???

    I believe this information about Oracle in a cluster is wrong. 

    I have talked with at least six different Oracle technicians and
    they swear, cross my heart, etc. that Oracle DOES NOT work in a
    master-slave relationship.  It works (just barely) as one would
    expect except that V5 (and probably V6) invoke a substantial increase
    overhead in disk I/Os since the pages and/or rows must be marked
    so other nodes will know the database is being updated.
    
    After I complete the next V6/TPS benchmark (mid December), I'll
    tell you exactly how Oracle V6/TPS works in a CI cluster.
    
    Where is this master-slave stuff coming from anyway ??

    
    Gerry 

>    If the node that is managing the ORACLE database goes down, the        
>    remaining users lose access to the database, (no failover), Correct???


Oracle 6.0 is supposed to have failover.

--david

    Sorry, about the master-slave stuff. It was something the ORACLE
    rep was telling the customer to impress him. I just wanted to find
    out how it really worked, since it didn't use the DLM.
    
    Thanx, for you responses, Paul