Tab: Indexing Performance

BATCHPARA

Number of processes for background parallel processing of initial indexing

Number of parallel background processes during initial indexing: the InfoCube's F and E fact table are split into a number of blocks specified in the BACTHPARA parameter. (The E table is split using the partitioning characteristic or another time characteristic. The F table is split using the request.) These blocks are then read from the database and written to the BW accelerator separately. (Note that a number of dialog processes, as specified in the NUMPROC parameter, are used for each background process.)

This applies to initial indexing only and not to roll up.

You can change the values for these parameters in transaction RSBATCH (process type TREXINDEXP).

03

NUMPROC

Number of Processes for Parallel Processing Using aRFC Dialog Processes During Indexing

The data is read from the database table on a package-by-package basis using Open Cursor and Fetch. The system calculates the package size from the width of the table and the default value for the package size (in bytes) in the system.

If you set a degree of parallelization that is greater than one, each of these packages is indexed in a new asynchronous job in dialog mode. When the indexing job for the package is started, the system reads a new package immediately. Ideally, you should reduce the indexing time to the time that is required to read the data and pack it for the RFC module.

If the degree of parallelization is equal to one, the system performs serial processing.

The optimization of the BW accelerator index cannot be parallelized in BW. However, optimization automatically runs in parallel on the BW accelerator server if the index is split.

5

RFCGROUP

LOGON Groups for the aRFCs

By specifying a LOGON group, you can determine which application servers the asynchronous dialog processes can start on. To do this, define a LOGON group and run program RSDDTREX_ADMIN_MAINTAIN with Object = RFCGROUP and Value = <LOGON Group Name>.

PKGSIZE

Package Size in Bytes for Internal Tables During Indexing Using aRFCs

100.000.000

SUBPKGSIZE

Package Size (Rows) for Export to Buffer During Indexing Using aRFCs

During dialog parallel processing with the NUMPROC parameter, packages with size PKGSIZE are read and indexed using aRFCs. To do this, the data packages must be transferred using Export to Data Buffer. This is done on a package-by-package basis since it requires a large amount of CPU and memory resources. The SUBPKGSIZE parameter specifies the number of rows for the package. We recommend a value between 10,000 and 20,000.

20.000

PARASIZETHRESHOLD

Number of rows after which a table is split into batch processes

The default setting is one million rows. With smaller tables, it is normally not worth the effort splitting it and starting multiple batch processes.

SPLIT_TYPE

Split Mode for the Batch Processes

Note that the split mode for an InfoCube's fact tables can be set for each InfoProvider during initial indexing. For more information, see Tab Page: Indexing Performance.