Skip to Content
author's profile photo
Former Member

Info Cube Partitioning vs. Multi Provider and timesliced cubes

We are working on an Info Cube that will have about 100 million rows in the fact table. Since this is quite a large number and 2/3 of that data is historical data that won't change any more after the initial load there are two design options: 1. partitioning of the info cube along 0CALMONTH or 0CALYEAR or 2. building separate info cubes for each time slice and join them in a multi provider.

Any suggestions or comments on the pro and cons of those alternatives would be much appreciated.

Frank Schülke, Ruhrgas AG - Germany

Add comment
10|10000 characters needed characters exceeded

  • Follow
  • Get RSS Feed

2 Answers

  • author's profile photo
    Former Member
    Posted on May 27, 2004 at 02:51 AM

    Hi,

    Both approaches may hold value. Many customers will partition by, for example, 0CALMONTH but also have base cubes for each year. You may want to look for some other logical partitioning criteria for the MultiProvider, if possilbe (like region, company code, etc). You may gain more parallel processing advantages with this technique. See OSS note 629541 MultiProvider: Parallel Processing. In short, the best approach generally is a combination of table partitioning (if available) AND multiprovider logical partitioning. Ron Silberstein, SAP - NW RIG (BI)

    Add comment
    10|10000 characters needed characters exceeded

  • Posted on Jun 08, 2004 at 04:08 AM

    Hello Frank,

    Ron is right. In the MultiProvider queries you can also restrict the virtual characteristic "InfoProvider" to read only current or historical InfoCubes.

    Regards,

    Marc

    SAP NW RIG US BI

    Add comment
    10|10000 characters needed characters exceeded