Conference ulysse::rdb_vms_competition

    Don't you believe any of this.  Obviously you didn't run any benchmarks
    in a cluster or the Oracle person couldn't possibly "bet his job"
    on this.  Ask the Oracle salesman for cluster references ... our
    customer and his outside MIS consultants have been trying for 6
    months in the US .... good luck.
    
    In benchmarks that I have performed using Oracle 5.1.22, the disk
    i/o rate tripled when I ran Oracle in a cluster (vs single cpu).
    As of last Sunday, Oracle V6.0.24 DOES NOT EVEN WARM START IN A
    CLUSTER  .... period !!!  Oracle V6 performance gains are greatly
    out of sync with reality.  Any actual performance gain in V6 is
    simply because the locking granularity is now at the row (vs table
    in V5.1.22).  In my benchmarks, I acheived the same run times between
    Oracle V5 and V6 when running a single program updating and inserting.
    I'm not so sure there are any performance gains in the Oracle engine
    itself.
    
    Re .3 I guess a question I have is whether or not you need to be faster
    than Oracle OR to simply show that RDB can do the job.  If you get
    sucked into the ".. as fast as Oracle ..", you'll be working at
    a disadvantage because Oracle will have succeeded in positioning
    itself as the performer to match/beat.
    
    Re: 2.  Since you're running on the 8820, my guess is that this
    system is not in a cluster.  FYI Oracle V5.1.22 will run on this
    machine but does not take advantage of SMP effectively.  Essentially,
    the 8820 will be running as though it is an 8810.  On the other
    hand, RDB will do just fine.  
    
    I don't think the ACMS thing is a matter of "supporting" Oracle
    and believe it is more along the lines of it is something that can
    be done.  I would have a hard time believing that Oracles forms
    were better than TDMS or VAXforms.
    
    Re .4  Those surveys don't mean squat.  In my way of thinking, Oracle
    is asking your customer to believe a source of data that doesn't
    even remotely have anything to do his business need.  Hopefully,
    your customer will be able to see through this.
    
    Try the old cost of ownership.  If the customer purchases Oracle
    V5.1 and goes on maintenance, he WILL NOT get the TPS option with
    Oracle V6.  The next product available for shipment from Oracle
    (IF they can ever get it ready) will be V6/TPS   ... NOT ... plain
    V6.  And oh, by the way, the TPS is about a 50-60% uplift in the
    states.  To make matters worse, "regular" V6 WILL NOT HAVE ROW LEVEL
    LOCKING.  The last number's I heard was that Oracle V6/TPS will
    cost approx $150,000 ... have your customer try that on for size.
    
    Hope this helps ... 
    
    --gerry
    
    
    

It is hard to follow Gerry's reply, but ...


I have posted a few notes in this file about Oracle's benchmark
methodology and the differences between what Oracle reports and
what we have reported.  Please read these notes.  

--david

    I'm sorry Dave, I got carried away .... sorta' like a shark's feeding
    frenzy .... I just couldn't stop rambling .... I'll do better next
    time.

    Gerry,
    
    There's a throw-away comment at the end of your note about 'regular'
    v6 not having row-level locking; you have to buy V6/TPS to get it.
       
    Did I understand that correctly? Am I alone in not realising this
    to be the case?
       
    So the correct answer to 'Will you have row-level locking' is not
    'yes' but 'only if you want to shell out another $150k'.
                                  
    Devious bunch, aren't they?   
                                  
    PDB                           
                                  

    
    
    The TPS option will add another $75,000 to the $125,000 DBMS price
    tag. Maintenance is 15% a year. Add it up. It is scary.
    
    Does anybody have any idea how DataPro 'surveys' a vendor's customer
    base. I do. DataPro calls up the product manager at the vendor to
    be surveyed and asks him/her for a list of customers (about 50)
    that DataPro could survey. The product manager gets a free copy
    of the results when finished. 
    
    Now what would any product manager worth his salt do in such a
    situation? Just what I did. I solicited the sales force nationwide
    to send me 2 or three of their very best customers. I then paid
    for the sales rep to take each of the selected customers to dinner 
    to 'explain' the significance of the survey that they were about
    to get inn the mail. I then sent each of these customers a letter
    from the CEO thanking them for their continuing support of the
    products...'please feel free to call me directly if there is anything
    that I can do.' 
    
    Needless to say, the results from 'my' DataPro survey were the best
    the company had ever received. Amazing. Just think what marketing
    animals like Oracle could do.
    
    Jamey

    The comment on ACMS in Europe is spurious. More than half of the
    ACMS licenses have been sold in Europe. There are many ACMS/Rdb
    sites there.
    
    ---- Michael Booth

    On the 9th December in London Codd and Date ran a one day seminar
    that evaluated a functional database benchmark on 14 major databases.
    
    I will not cover the detail of the event ( see Chefs::VIA_FORUM
    note 155 for details) however the press coverage has been very
    interesting.
    
    On the front page of DEC Computing 14th December "Poor Result for
    Oracle in DB Report" - the article starts "Relatives of the Clanton
    Gang, who were gunned down at the OK Corral by Doc Holliday, Wyatt
    Earp and friends , probably know how Oracle is feeling at the moment"
                 
    It goes on to say that Oracle came out the worst of all.
    
    On page three the full details are printed with the heading "RDB
    WINS HIGH PRAISE FROM THE LEADING DATABASE GURUS".  Within the body
    of the article it states "that on the whole results were encouraging
    for almost everyone - except Oracle, which failed to complete the
    test".
    
                
    

	We have had benchmark-competition between ACMS/TDMS/Rdb/SQL-Cobol 
	and Oracle-applications during last two weeks. 

	We had identical applications. We used ACMS/TDMS/Rdb/SQL-Cobol
	and they (=one software house, not ORACLE) used Oracle.

	The software house defined the application and database (=tables 
	and indexes). After that we had to implement identical applications
	and load our databases with same data. 

	The test transaction consisted three separate forms. 
	- Form 1: We read information into a form from three different 
	  	  tables (one record from two tables and 10 records
		  certain order from 1 table). Display only form.
	- Form 2: We read 10 records certain order from one table 
		  and 10 records also certain order from another 
		  table. Display only form. After form two we read next
	          free order number from parametertable and updated
	          that record.
	- Form 3: We inserted 10 orderlines into database. With each
		  order line we read one record from 5 different tables,
		  we updated one record and inserted one orderline.

	We used Rdb V3.0 and we used multi-file. We had five different 
	areas (+ one area for system relations). We used three disks
	in our tests (one disk was system disk RA81). Our database 
	located in two other disks (2*RA81).
	
	We both ran tests on MicroVAX 3600. We tuned our VMS-parameters
	and also ACMS-parameters.

	The Software House wanted that we inserted same 10 products. So
	we had to update ordered amount into same records (=sometimes
	little waiting time).

	The measurement was which one could insert more orderlines
	during ten minutes. We had 14 end-users in our tests.

	- We inserted max. 1316 orderlines. We had also one test where were
	  end-users from the customer and Software House. Then we inserted
	  1281 orderlines.

	- The Software House with their Oracle-application inserted 800
	  orderlines. They didn't show us how they run tests. But they
	  didn't use end-users, they read inputdata from a file.

	We implemented also another version, where we inserted 10 order 
	lines in one time. With that version we inserted 1523 order lines/
	10 minutes. We had big differences between end users. One inserted
	140 orderlines and some only 80 orderlines.

	The Software House with their Oracle-version with their used
	whole CPU. So they couldn't more lines than 800.

	Our problem was disk-I/O. But we are sure that our amount 
	1523 is not any maximum limit. 

	The customer hasn't yet decided which one they choose.

	Mervi

	The customer decided this week and they choosed us. They are
	going to implement their applications with ACMS/TDMS/Rdb/SQL/C
	and Rally.