[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

24.0. "Oracle questions from customers." by 51586::KEW (I'll let the fancy take you...) Mon Aug 24 1987 15:11

Hi folks,

Warning
this is from a conference being held at Australian DECUS, thus visible to 
customers. The conference is only on the net when DECUS people aren't 
there. I thought that viewers of this conference might be able to answer 
the notes here. I'm sure answers from the easynet are going to impress the 
hell out of these people.

Jerry

        <<< DUNDEE::DECUS_1:[NOTES$LIBRARY]INFORMATION_SYSTEMS.NOTE;2 >>>
================================================================================
Note 2.0                     ORACLE RDMS Performance                     1 reply
CROCKY::MILLERD                                       4 lines  24-AUG-1987 16:59
--------------------------------------------------------------------------------

    I would like to exchange information about the performance of ORACLE
    RDBMS on VAXes.  On our 11-750, response degrades unacceptably beyond 6
    to 8 users actively entering data via SQL*Forms.  Do other sites do
    better or worse than this? 

================================================================================
Note 2.1                     ORACLE RDMS Performance                      1 of 1
CROCKY::LIGTERMOETG "GERRY LIGTERMOET "               5 lines  24-AUG-1987 17:12
--------------------------------------------------------------------------------
                          -<  A future ORACLE user. >-

    We are about to install ORACLE on an 11-780. Performance is a concern.
    do have any more comments that may be of hlp to me.
    
    Gerry
    
T.RTitleUserPersonal
Name
DateLines
24.1ORACLE performanceDYO780::ROARKWed Sep 02 1987 12:4363
    The following is a condensation of a paper presented at the Oracle
    users conference (1986) by Greg Adams of Glenn A. Barber and Associates
    in Studio City Calif.
    
    observed practical limits for a moderate size Oracle application
             Processor      Number of Oracle Users,(physical memory-MB)
             ---------      ----------------------
             725/730          2 (with 3-5 MB)
             750              6-10 (with 6-10MB)
             Micro Vax II     8-12 (with 8-12MB)
             780              10-18 (with 10-15MB)
             785              15-25 (with 15MB or more)
    
    The rule of thumb developed is there should be 1M of memory for
    the first 12 users and 1/2M of memory for each user after that.
                               
    Once there is sufficient physical memory other VMS parameters
    can be adjusted to get better performance out of the system.
    
    The paper reccomends the following values based upon the observation
    that large IAP applications require 2000 - 3000 pages.
       WSDEFAULT 200 TO 500 pages
       WSQUOTA   1000 to 2000 pages
       WSEXTENT  2000 to 5000 pages  use Authorize to change.
    
       WSINC = 10% to 20% of the diference between the average wsdefault
    and wsquota. ~150 to 250 to correspond to the above numbers.
       WSMAX  should ber as large as the largest extent authorized 
       WSMAX and WSINC should be adjusted using SYSGEN OR AUTOGEN

    It uses system buffers and uses 8K segments so MAXBUF should be set to 
    8192 bytes.
    This parameter should be set using AUTOGEN or pool starvation and
    other nasty things could occur on loaded systems.
    
    Other things that can be done are to increase the number of buffers
    in the init.ora and adjust buff_hash_bkts so the ratio of
    buffers falls between 4 and 8, the max value of buff_hash_bkts is 32.
    buff_hash_bkts is going to affect length of search list.
    Adjusting init.ora parameters can affect the size of the ORACLE SGA(system
    global area) increasing the number of global pages needed.  We cut
    the disk IO by 50% 42 /sec to 20/sec by just increasing buffers and
    buff_hash_bkts.  Be sure you link ORACLE at the beginning with an ample 
    SGA size. When the size of the SGA outgrows the size you've linked ORACLE 
    to originally, all the applications have to be relinked as well as ORACLE 
    of course if you're going to increase parameters like buffers.  You don't 
    need     to use all of the SGA size that you've linked ORACLE to.  We've got
    86 and 8700 and 780's in a cluster with a shared database.  The machines
    with more memory have a larger installed image (higher values of buffers,
    processes etc in the init.ora for those nodes, while the 780's use
    smaller values with the smaller memory.
    Placing the before image file and database file on different disks
    can help, particularing in update active environment.  Also in
    the FROM clauses on joins place the tables in decreasing order by 
    number of rows,eg. FROM  table_with_most_rows,some_rows_table,fewest_rows_
    table.  Make sure the queries make goo use of indexes, use noncompressed
    indexes if you can afford the space.  Constructing good space definitions
    will help keep the too many extents (internal to oracle) from happening.  
    Be aware of operator precedence in constructing queries.
    
    --Tim