Definition of Clustering

Use

Prerequisites

You can only change clustering if the DataStore object does not contain any data. You can change the clustering of DataStore objects that are already filled using the Reclustering function.

For more information, see Reclustering.

Features

In the DataStore maintenance, select Start of the navigation path Extras Next navigation step DB Performance Clustering End of the navigation path .

You can select MDC dimensions for the DataStore object on the Multidimensional Clustering screen.

Select one or more InfoObjects as MDC dimensions and assign them consecutive sequence numbers, beginning with 1. The sequence number shows whether a field has been selected as an MDC dimension and determines the order of the MDC dimensions in the combined block index.

When selecting, proceed as follows:

Select InfoObjects that you use to restrict your queries. For example, you can use a time characteristic as an MDC dimension to restrict your queries.
Select InfoObjects with a low cardinality. For example, the time characteristic 0CALMONTH instead of 0CALDAY.

You cannot select more than three InfoObjects.
Assign sequence numbers using the following criteria:
- Sort the InfoObjects according to how often they occur in queries (assign the lowest sequence number to the InfoObject that occurs most often in queries).
- Sort the InfoObjects according to selectivity (assign the lowest sequence number to the InfoObject with the most different data records).
  
  Caution
  Note: At least one block is created for each value combination in the MDC dimension. This memory area is reserved independently of the number of data records that have the same value combination in the MDC dimension. If there is not a sufficient number of data records with the same value combinations to completely fill a block, the free memory remains unused. This is so that data records with a different value combination in the MDC dimension cannot be written to the block.
  
  If for each combination that exists in the DataStore object, only a few data records exist in the selected MDC dimension, most blocks have unused free memory. This means that the active tables use an unnecessarily large amount of memory space. Performance of table queries also deteriorates, as many pages with not much information must be read.

Example

The size of a block depends on the PAGESIZE and the EXTENTSIZE of the tablespace. The standard PAGESIZE of the DataStore tablespace with the assigned data class DODS is 16K. Up to Release SAP BW 3.5, the default EXTENTSIZE value was 16. As of Release SAP NetWeaver 7.0, the new default EXTENTSIZE value is 2.

With an EXTENTSIZE of 2 and a PAGESIZE of 16K, the memory area is calculated as 2 x 16K = 32K, which is reserved for each block.

The width of a data record depends on the width and number of key fields and data fields in the DataStore object.

If, for example, a DataStore object has 10 key fields, each with 10 bytes, and 30 data fields with an average of 9 bytes each, a data record needs 10 x 10 bytes + 30 x 9 bytes = 370 bytes. In a 32K block, 32768 bytes/370 bytes could write 88 data records. At least 80 data records should exist for each value combination in the MDC dimensions. This allows optimal use of the memory space in the active table.