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

Conference smurf::ase

Title:ase
Moderator:SMURF::GROSSO
Created:Thu Jul 29 1993
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:2114
Total number of notes:7347

1892.0. "1000+ DRD's?" by USPS::FPRUSS (Frank Pruss, 202-232-7347) Thu Feb 20 1997 20:01

    Cross Post to Oracle:
    
    I have been talking to the DBA of the customer who is buying a
    TruCluster to support Oracle and Oracle Financials.
    
    He is going through a worksheet and some Implementation guides and is
    coming up with about 1 Tb and over 1000 unique tables.
    
    Now, from what I hear, each Oracle table will get at least one DRD (the
    maximum size of DRD that Oracle use being 14 Gb).
    
    So we would talk about a TruCluster with over 1000 drd's
    
    Has anyone run across any Oracle TruCluster configuration, either
    proposed or sold that looks anything like this?
    
T.RTitleUserPersonal
Name
DateLines
1892.1And any tools / scripts out there?USPS::FPRUSSFrank Pruss, 202-232-7347Thu Feb 20 1997 20:024
    And do we have any tools to automate creation and sizing of 1000 OF lsm
    virtual units, plexes and DRD's?
    
    FJP
1892.2And what about SAP?USPS::FPRUSSFrank Pruss, 202-232-7347Thu Feb 20 1997 20:383
    Just talked to a benchmark center guy & sez SAP uses 9000 TABLES.
    
    Hmmm...
1892.3some commentsALFAM7::GOSEJACOBFri Feb 21 1997 07:4342
    re .0
    >He is going through a worksheet and some Implementation guides and is
    >coming up with about 1 Tb and over 1000 unique tables.

    I have no idea what the calculations in that sizing guide look like.
    One of the questions to ask is how the sizes add up to 1 TB; e.g. is
    this something like: you'll have these 1000 tables and they all be
    around 1 GB if you have 100 000 customers. So what I'm driving at: the
    calculations may become inacurate the bigger the application gets.
    Keep in mind: this is just wild speculation on my side.

    >Now, from what I hear, each Oracle table will get at least one DRD (the
    >maximum size of DRD that Oracle use being 14 Gb).

    Where did you get that from? I mean the table/drd mapping. Two comments
    here:

    1) Yes there are limits within Oracle for the maximum size of a
       datafile. I can't remember the exact formula but the max datafile
       size depends on the Oracle block size. And the absolut maximum of
       14GB sounds about right.

    2) Oracle tables are allocated in so called extends. One table can have
       several extends up to a limit. All extends of one table must be
       allocated in one tablespace (this limitation exists in Oracle7). A
       tablespace can hold any number of tables. A tablespace comprises 1
       or more datafiles of any kind (raw, UFS, AdvFs) and size.
     
       So  there may be a recommendation to put each table in a seperate
       tablepace for performance reasons but there is no strict rule to
       that and its also a pretty unusual approach with any larger number
       of tables.

       As an approch to the other extreme you may decide to throw all
       your drds into one single tablespace and allocate extends for all
       1000 tables there. 

    So if you have to build a 1 TB database you only need 74 drds (74 *
    14GB > 1 TB). You may want to use more for better control over which
    table goes onto which disk.

    	Martin
1892.4perl it!LEXSS1::GINGERRon GingerFri Feb 21 1997 12:3116
    As to tools for building large configurations, I have written a couple
    perl scripts that do the hsz config and the LSM config. I create a
    table  of the basic units and these scripts issue all the needed
    commands.
    
    I did this as a part of a disaster recovery plan. At a disaster site it
    is unlikely you will have identical hardware config as your home site.
    And sitting at a disaster site dragging around a mouse on a gui did not
    strike me as efficient use of the allowed time. So all we need to do is
    look over our hardware, re-write a couple control tables and run the
    scripts. I can re-build our 16 HSZ's and all LSM structures in 5-10
    minutes.
    
    Sorry, but the scripts are not generic, they have many assumptions
    built in that match our site, so they would be of little use to anyone
    else. But you should be able to do the same thing for any site.