RE: Moving database from HP to IBM-AIX

  • From: "Goulet, Dick" <DGoulet@xxxxxxxx>
  • To: <dubey.sandeep@xxxxxxxxx>, "Mark W. Farnham" <mwf@xxxxxxxx>
  • Date: Fri, 28 Oct 2005 12:16:34 -0400


        If you want to use Pro*C you can have multiple connections to
multiple databases at one time and use array processing.  The problem is
that your program will be peculiar to the application. 

-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx
[mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of Sandeep Dubey
Sent: Friday, October 28, 2005 12:02 PM
To: Mark W. Farnham
Cc: oracle-l@xxxxxxxxxxxxx
Subject: Re: Moving database from HP to IBM-AIX


Thanks for the reply.

I am thinking on ways to optimize data unloading and transfering.

1. Is it possible to use pipe so that as data is unloaded from source
table, ftped and is consumed by sqlloader to pump in?

2. Is there any faster tool to unload the data from table to flat
file. Will Pro C be faster than sqlplus spool?

3. How big should I have a flat data file. 4 GB enough or I can have
bigger file size.?

4. How best to use parallel processing?

Thanks again


On 10/28/05, Mark W. Farnham <mwf@xxxxxxxx> wrote:
> This is usually a balancing act between "death by details" and
> on re-creation of indexes on your largest objects. It is very good
that you
> know your target window, so you can execute a test with a minimum of
> an see if that is good enough.
> If the full export/full import runs in under 24 hours, why worker
> Unless you're running a MAID disk farm you probably don't even save
> electricity.
> Let's assume for the moment though, that is not fast enough.
> I'm not sure whether you have a full clone testbed of your warehouse
> which to get a testing source, but even if you don't, getting a dump
of what
> is for testing will be well worth a short additional outage so you can
> Interim changes for a warehouse between the test runs and the real cut
> should not be material.
> So, you get a full no-rows export for starters. Get the data exports
by user
> and/or by user/table leaving out the big ones. Generate create table,
> trigger and constraint scripts for the big ones.
> Unload the data from the big ones in a format that loader will swallow
> Consider whether there is a predominant order of access supported by
> index for any of these big guys that might be worth the trouble to
> and if there are any serious outliers and especially with partitioning
> you could load in parallel, you might want to also unload in pieces.
> said warehouse, so I'm guessing your rows no longer change in length,
so if
> you're not already aggressive with percent free you might want to get
> aggressive with a small percent free. If older stuff is row length
> and younger stuff is not, consider unloading pieces by age so you can
> the older stuff denser and then adjust percent free for the younger
> You *may* benefit from ordering the unload if there is either a
> pattern of access that will cause corelation between block selectivity
> row selectivity in future use of the database, or if it will allow you
> build an index using the already sorted option.
> Run the no-rows import. Muck around and disable what you need to so
you can
> start the big boys loading in parallel streams (separate runs of
> to a reasonable load average on your CPUs rather than parallelism in
> so you don't pay the slave co-ordination overhead). As each big boy
> finishes, and taking into account max IOPs on your temporary space and
> reasonable load average on your CPUs, start indexing it. Again, you
> win by multiple sqlplus sessions over parallelism.
> Good luck, and don't do more work than you have to.
> ------------------------------------
> Rightsizing, Inc.
> Mark W. Farnham
> President
> mwf@xxxxxxxx
> 36 West Street
> Lebanon, NH 03766-1239
> tel: (603) 448-1803
> ------------------------------------
> -----Original Message-----
> From: oracle-l-bounce@xxxxxxxxxxxxx
> [mailto:oracle-l-bounce@xxxxxxxxxxxxx]On Behalf Of Sandeep Dubey
> Sent: Friday, October 28, 2005 10:41 AM
> To: oracle-l@xxxxxxxxxxxxx
> Subject: Moving database from HP to IBM-AIX
> Hi,
> We need to move 2 TB data warehouse on Oracle 9.2 HP to Oracle 9.2
> database on AIX. These operating systems has opposite endian. Down
> time should be less than 24 hrs.
> Please let me know what will be faster and optimum way to do this. I
> am thinking of sqlloader as the fastest way to go. Are there any
> better alternative?
> If you have undergone such exercise please share your lesson learnt.
> Regards
> Sandeep
> --


Other related posts: