Conference ulysse::rdb_vms_competition

>    In previous topics, the fact that Oracle performs poorly in a cluster
>    is mentioned. The memory buffer becomes a problem. My question is
>    	 	Does Oracle 6.0 work at all in a CI Cluster ?
>    		If so, are there any benchmark numbers available for
>    		cluster performance? E.g. 6240 compared to a 2x6220 ?
                
Oracle 6.0 does work in a CI cluster (as well as LAVC).  Previous
versions worked too.  The traditional problem has been supporting
after image journaling - Oracle didn't on clusters.  They have
reportedly removed this limitation.  Realize, however, Oracle 6.0 is
not released yet.  When it is, you will have to purchase Oracle 6.0 with
TPS (60% more in license fees).  Oracle 6.0 will not be available for quite
a while without TPS.

Oracle has not released any benchmark numbers that compare cluster and
non-cluster performance.

Informal talks with Oracle engineers have revealed that there are significant
performance hits over a cluster.  These findings have been discussed in other
topics - look there for detail.

In a nutshell, the reasons for the large performance difference is that
non-clustered systems avoid using the VMS lock manager and can take
advantage of large shared memory.  Fast commit strategies work fairly
well in this type of environment where I/O is deferred to a later time
(usually till after the benchmark is completed).  In a cluster environment
you must use the lock manager and you don't have shared memory over the CI
bus.  Thus, locking (CPU), contention, and I/O all increase.

The buffers in shared memory must be written back to disk each time the
other node in the cluster wants a page.  ASTs and value-block locking
must be performed to signal the writing of the buffers.  Performance
falls as a result.

Compare this to Rdb where the system has been designed to work in a 
clustered/networked environment.  The monitoring, tuning, and physical
design control that Rdb gives you, together with other products like
ACMS and TDMS also allows the user to have a high degree of performance
and security on single node machines.

Don't let Oracle's benchmark results fool you.


Hope this helps.  Call if you have questions.

--david
381-2893

    I have run standalone benchmarks with Oracle V5.1.22 and Oracle
    V6/TPS.  Sunday, I will be running the V5 cluster benchmark.  On
    Dec 8-10, I'll be running V6/TPS cluster benchmark.  These are all
    CI class machines (8820, 8840, 6240) so this isn't small stuff.    
    I'll do my very best to post results here by Jan 1.
    
    Until then, David's right.  Don't let those numbers fool you and
    don't let Oracle tell your customers that 6240 will outrun an 8840
    much less an 8820.  In my benchmark, 6240 was approx 30% SLOWER than
    the 8820.  Finally, and trust me on this .... we didn't get into
    double digit TPS on the 6240 with a relatively large database. 
    Naturally a key concept here is what you call a TPS so it's always
    open to one's imagination.
    
    Call me if you want to talk.  DTN 452-3445.
    
    --gerry

    Thanks both of you for the info. Tried to call but couldn't use
    your DTNs from Sweden.
    
    However, after having asked the question, but before having read
    your answers, Oracle Sweden told us that 6.0 is NOT supported on
    clusters because of lack of a common LOG function. This is getting
    more and more confusing.
    
    To give you some background, we are in a competetive situation with
    NCR and Pyramid for a system to run Oracle on. I.E. the customer
    has already decided to use Oracle and there's nothing to do about
    it. We have sold the cluster idea for security reason to the customer
    and are now in a bit of a tricky situation. 
    
    Any more info?
    
    Daniel 

>                     -< Oracle claims it's not supported >-

Has ORACLE Sweeden contacted ORACLE USA?  On this side of the pond
Oracle continues to talk about their support for the VAXcluster.

In release 5.* of Oracle there was no AIJ (after image journaling ==
rollforward journaling) on a VAXcluster.  This is supposed to be
fixed in 6.0.  Performance, however, is another issue and has been
discussed before.

--david

    We tried to load V6/TPS on a cluster Sunday.  The database would
    not come up "shared", which is an Oracle requirement before you
    can bring up V6/TPS on the other node(s) in the cluster.  When the
    Oracle person tried, an ORA-600 error was shown.  Basically, the
    600 message is a catch all when something went wrong that Oracle
    was unable to trap.  (FYI this was V6/TPS version labeled "6.0.24.1.0
    - Limited Production" which my customer received on Dec 7.)
    
    I too am in the position where the customer has chosen Oracle. (They
    have Honeywell and VAX now.)  Needless to say, the customer and
    the Big 8 accounting firm consultant are very hot since an area
    Oracle VP told them late last Friday that V6/TPS would cluster just fine
    "but may not perform well".  It seems that literally all the sales
    force and most of the higher level sales support mgrs at Oracle
    were believing V6/TPS would cluster.  They all got a good "hood
    winking" and now know how their own customers feel.
    
    My customer and consultant are having a non-disclosure with Oracle
    in Belmont this morning.  I should be able to find out a few details
    of this meeting in a day or two.  
    
    In the V6 VMS installation manual, Chapter 8 describes how to install
    V6/TPS on a VAXcluster.  Also, Chapters 21 and 22 of the V6 DBA
    guide describe multiple instances and databases on different CPUs.

    If you want to talk, call me direct at USA 816 943 3445.
    
    --gerry    

    Assumptions: 
    	- Oracle is primarily successful in its marketing and sales
	  based on the message that it is the only viable multi-platform
	  Relational DBMS
    	- Oracle has achieved the bulk of its licenses and revenues from
	  VAX (presumably VMS) sales
    	- Oracle V5.0 and V6.0 do not play well with the VAX/VMS strengths
	  (clustering, and presumably networking)
    
    Conclusions: The MARKETING MESSAGE has not, in fact, been the
    end-result... ORACLE got customers to THINK they needed A, when
    in fact all the customer wanted was B, and then carefully delivered
    a sub-standard B.
    
    Would it be viable to point out to customers who experience ORACLE
    inadequacies on VAXclusters that:
    
    	- VMS is where ORACLE has the largest installed base... if the
    software works like that on their most successful platform, what's
    it going to work like on secondary (DG, Prime, whatever) platforms?
    
    	- the VMS version is what ORACLE delivers FIRST... secondary
    platforms are brought up to latest rev later... possibly 6-12 months
    later (I don't know the average or longest delays experienced).
    
    	- Why not go with the flow and accept that a jack of all trades
    is master of none, and get the DBMS designed for the system: we
    can easily outsell the second tier companies, and we can win against
    IBM in most cases.  I don't even mind if IBM gets another DB2 license
    to replace the ORACLE/MVS license.
    
    Steve