Dynamic Precision BPC 7.3 Instrukcja Użytkownika Strona 9

  • Pobierz
  • Dodaj do moich podręczników
  • Drukuj
  • Strona
    / 19
  • Spis treści
  • BOOKMARKI
  • Oceniono. / 5. Na podstawie oceny klientów
Przeglądanie stron 8
Using SQL Server 2012 Column-Store with SAP BW
SAP COMMUNITY NETWORK scn.sap.com
© 2012 SAP AG 9
Aggregates
By creating a fully optimized aggregate, you can enormously speed up the runtime for a particular BW query.
However, customers often try to avoid the overhead of creating BW aggregates. If aggregates are really
needed, customers try to find a compromise between the overhead of aggregate maintenance and BW query
performance: A single BW aggregate is typically not fully optimized for a single BW query. In return, the
same aggregate can support a variety of BW queries at the same time.
When using the column-store, you typically do not need aggregates at all. Therefore, you should deactivate
the aggregates of a cube at the same time when defining the column-store (in report MSSCSTORE, see
below). The definition of the aggregates then still exists in SAP BW. However, the database tables of the
aggregates are dropped.
Note: We recommend not using BW aggregates for cubes using the column-store for the following reasons:
1. BW aggregates never use the column-store. This is by design in to support the BW Change Run
2. The BW OLAP engine is not aware of the column-store. Therefore, the OLAP engine may decide to use
an aggregate although the basis cube has a column-store index
3. Typically, queries using BW aggregates are slower than queries using a column-store index
4. Not using aggregates results in further disk space savings and less administration efforts
BW compression
Since the non-clustered column-store index is read-only, it has to be dropped at the beginning of the BW
cube compression and re-created at the end of the cube compression. This is all performed automatically. A
BW administrator has not to take care of this, with one exception: If the BW cube compression fails for any
reason (for example due to a full database transaction log), the column-store index may not exist anymore
on the database. In this case, the index has to be created manually using SAP transaction RSA1 (see below
at “Repair indexes”).
You may argue that the BW cube compression takes more time when using the column-store, since you
have to re-create the column-store index. However, it is often the other way round. You do not need to
maintain all the single column indexes of the e-fact table when compressing a cube with column-store index.
The index creation of a column-store index is very fast. By default, SAP BW is requests eight SQL Server
threads for this task. You can further increase the parallelism for index creation by setting the RSADMIN
parameter MSS_MAXDOP_INDEXING (see below).
The BW cube compression on Microsoft SQL Server is today much faster than a few years ago. By using the
SQL MERGE statement, the aggregation from the f-fact to the e-fact table has become faster. The deletion
of the rows from the f-fact table has become much faster due to an improved partition drop algorithm in SQL
Server 2008 SP2 and newer SQL Server releases.
However, the cube compression nowadays contains an additional step. It runs an update statistics on all
database tables of the compressed BW cube. You can turn off the update statistics by setting a RSADMIN
parameter. However, we do not recommend doing so. For details, see
http://blogs.msdn.com/b/saponsqlserver/archive/2012/08/20/how-update-statistics-really-works-in-sap-
bw.aspx
Note: BW query performance is best, if all BW request are compressed and therefore the column-store index on the e-
fact table can be fully leveraged. However, you do not need to compress a cube each time a new request is loaded
into the cube. You can schedule BW cube compression once a day or even once a week. Thereby you make sure
that the majority of the BW request is always compressed. This is typically sufficient to get a good BW query
performance.
Przeglądanie stron 8
1 2 3 4 5 6 7 8 9 10 11 12 13 14 ... 18 19

Komentarze do niniejszej Instrukcji

Brak uwag