[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference ulysse::rdb_vms_competition

Title:DEC Rdb against the World
Moderator:HERON::GODFRIND
Created:Fri Jun 12 1987
Last Modified:Thu Feb 23 1995
Last Successful Update:Fri Jun 06 1997
Number of topics:1348
Total number of notes:5438

535.0. "oracle poor performance" by MSAM01::LIMYAUTONG (desperado......) Mon Jan 15 1990 09:01

    I have a customer running Oracle on the MV3400 with extremely poor
    performance. Is there anyway of convincing the customer that it is the
    Oracle rather then our hardware that is giving the problem. Any help
    will be appreciated.
T.RTitleUserPersonal
Name
DateLines
535.1Need more infoKCBBQ::DUNCANOracle ... the designer databaseWed Jan 17 1990 03:1335
�� I have a customer running Oracle on the MV3400 with extremely poor
�� performance. 

Gosh, except for the local account with a Cray, I've never seen a system where
the customer was happy with Oracle performance. 

Need to define what "poor" means in reference to a previous point when the
system performed OK. For example, did the MV3400 start performaing bad AFTER
Oracle was loaded, after 50 users logged on, after ______ ???  Also, how was the
"poor" performance measured ? Batch throughput, terminal response time, VAX
monitor, etc. 


�� Is there anyway of convincing the customer that it is the Oracle rather then
our hardware that is giving the problem. 

Only with numbers from VPA, SPM, or VMS monitor.  Anything else is likely to be
emotional and/or higly inaccurate.

Why should we have to convince them it is Oracle or the MV3400 at fault ? Is it
possible that the mv3400 is sized wrong ? Why do they think it is the hardware ?
What can't the problem be Oracle ? Can they justify their statements with
numbers ? 

�� Any help will be appreciated. 

I can probably help you but need to know a lot more detail.  Start with getting
the customer to tell you how the system (hardware and software) was sized and
what assumptions they used.  Then get them to tell you EVERYTHING they have done
to improve performance, what kind of discussions they have had with Oracle, with
Digital, and with themselves.  Then give me a call or send Email and I'll try to
help.

-- gerry

535.2a war storyWILARD::DELLISOLAit blinded me with scienceFri Jan 19 1990 17:1357
    What I usually do is to check the VMS performance without Oracle 
    running. After I  establish the baseline, we fire up Oracle and
    measure some more.
    
    It's been my observation that whether Oracle was installed  by the
    customer or by the Oracle reps,  all of the default installation
    parameters were taken.  Worse yet, the database gets squeezed
    onto a couple of drives which were hard hit prior to Oracle.
    
    The Oracle folks usually are more application-oriented and it seems
    like they get nervous around the potential sizing of machines and
    VMS performance.  So the finger pointing about machine sizes and
    platforms and VMS.  It's VERY EASY to say "VMS can't do it" or "That
    VAX is too small."  If Oracle has entrenched itself onto your site,
    it makes it very difficult. .1 is correct.  Why should we have to
    prove that it's not our hardware or operating system?  If the customer
    is upset, have them request a performance study from both Digital
    and Oracle.  Make sure that you go face to face with the Oracle
    reps.  We have more experience with VMS and our hardware, we can
    certainly establish our credibility. Keep calm, stay technical and
    challenge them.
    
    Once, an Oracle rep said to me, "we have to reduce the number of direct
    I/Os to that drive.  The customer needs a faster drive.  It's a
    bottleneck. The application will run better if the databse is on
    a faster drive instead of the two drives which it's on."
    
    I said, "The drives are not fragmented, there is only one file, your 
    database. There are no split-ios, so the  XQP is able to satisfy 
    the requests.  I think the problem is with the application design and 
    how you've set up the read/write requests.
    
    Making this customer buy a faster drive is the wrong move because
    unless there is something really wrong (like fragmentation or lots
    of other requests), the performance really won't improve."
    
    I should mention that the data base was on 2 RA90s (not shadowed),
    isolated on their own HSC requestor. The problem was in the way Oracle
    wrote the SQL requests, not in the hardware performance. Also, it
    didn't help that they didn't load the database with the customer's
    newest data, so basically, the jobs were infinitely looping, looking
    for data which didn't exist!
                                   
    In this  war story, we came out ahead.  Future development with
    Oracle is on hold and the customer is writing an Rdb -Oracle position
    paper.  It has been noted that performance problems were not evident
    with the use of RDB based applications.
    
    Also, in 2 of this company's divisions, future development will be
    RDB based. The reason was that they believed that Digital was in
    a better position to provide support for applications and future
    capacity planning for the platforms.
    
    Regards,
    
    Kathy