Alan Hargreaves' Blog

The ramblings of an Australian SaND TSC* Principal Field Technologist

My CMT machine loads Oracle Databases slower than ..

This is more of an “Oh no not again” type post, …

I am constantly amazed at the number of escalations that make it to the performance group with this as the problem.

It really is a case of an unrealistic expectation and knowledge of what the machines excel at.

The most recent of these to cross my desk talks of a customer concerned that a dual core 2.5GHz x86 based box loads data into an Oracle database much faster than his shiny new T5220.

Until such a time as Oracle makes their SQL Loader run multi-threaded (which may bring in problems all of their own) this will always be the case.

The design of the system is such that it will run single threaded applications much slower than the x86 counterparts. These machines, however, come into their own once we enter production and start getting lots of parallel requests on the database. As we are running far more cpus, the load on the database must be much higher before we start to see any significant degradation.

The question that really should be asked here is, “Where do you want your performance? In the database load that you will do once, or in responding to production queries?”

Advertisements

Written by Alan

March 2, 2009 at 11:30 pm

Posted in Solaris

4 Responses

Subscribe to comments with RSS.

  1. There is an Oracle tool that lets you do the import on multiple threads: http://www.oracle.com/technology/products/database/utilities/htdocs/data_pump_overview.html

    Mircea Crisan

    March 3, 2009 at 12:57 am

  2. Apropos:
    since Sun Microsystems does not sell any sparc based systems which are able to compete on price with generic i86pc hardware, the Sun sales force is attempting to sell CMT systems as a panacea solution for database workloads.
    And as far as DataPump is concerned, DataPump is highly unreliable (I used to migrate Oracle databases for a living, day in, day out), and DP requires that the source DB be exactly the same as the destination DB in terms of revision.

    UX-admin

    March 3, 2009 at 7:31 am

  3. Wouldn’t it be relatively easy to write up a few sample scripts in Perl/Python etc. that simply splits the input file in a semi-intelligent fashion and pipes it to multiple import processes?

    patrick giagnocavo

    March 3, 2009 at 10:53 am

  4. Alan
    We’ve had similar issues replacing a V1280 with a T5220. The general system performance in the application is equal or better with the T5220. The only issue we’ve had is with a weekend batch job that is very single threaded -but it seems to only be a problem with the write-intensive parts of the job.
    I’m doing some database rebuilds – not with Oracle though – at the moment and haven’t noticed any real performance differences between the two platforms.
    Gamberoni

    Gamberoni

    March 4, 2009 at 3:04 am


Comments are closed.

%d bloggers like this: