Page 19 - DCAP603_DATAWARE_HOUSING_AND_DATAMINING

Unit 1: Data Warehouse Practice

1.6.4 oLap server architectures notes

We describe here the physical implementation of an OLAP server in a Data Warehouse. There are
three different possible designs:
1. Relational OLAP (ROLAP)
2. Multidimensional OLAP (MOLAP)
3. Hybrid OLAP (HOLAP)

roLap

ROLAP stores the data based on the already familiar relational DBMS technology. In this case,
data and the related aggregations are stored in RDBMS, and OLAP middleware is used to
implement handling and exploration of data cubes. This architecture focuses on the optimisation
of the RDBMS back end and provides additional tools and services such as data cube navigation
logic. Due to the use of the RDBMS back end, the main advantage of ROLAP is scalability in
handling large data volumes.

Example: ROLAP engines include the commercial IBM Informix Metacube (www.ibm.
com) and the Micro-strategy DSS server (www.microstrategy.com), as well as the open-source
product Mondrian (mondrian.sourceforge.net).

MoLap

In contrast to ROLAP, which uses tuples as the data storage unit, the MOLAP uses a dedicated
n-dimensional array storage engine and OLAP middleware to manage data. Therefore, OLAP
queries are realised through a direct addressing to the related multidimensional views (data
cubes). Additionally, this architecture focuses on pre-calculation of the transactional data into
the aggregations, which results in fast query execution performance. More specifically, MOLAP
precalculates and stores aggregated measures at every hierarchy level at load time, and stores
and indexes these values for immediate retrieval. The full precalculation requires a substantial
amount of overhead, both in processing time and in storage space. For sparse data, MOLAP
uses sparse matrix compression algorithms to improve storage utilisation, and thus in general is
characterised by smaller on-disk size of data in comparison with data stored in RDBMS.

Example: MOLAP products are the commercial Hyperion Ebasse (www.hyperion.com)
and the Applix TM1 (www.applix.com), as well as Palo (www.opensourceolap.org), which is an
open-source product.

hoLap

To achieve a tradeoff between ROLAP’s scalability and MOLAP’s query performance, many
commercial OLAP servers are based on the HOLAP approach. In this case, the user decides which
portion of the data to store in the MOLAP and which in the ROLAP. For instance, often the low-
level data are stored using a relational database, while higher-level data, such as aggregations,
are stored in a separate MOLAP. An example product that supports all three architectures is
Microsoft’s OLAP Services (www.microsoft.com/), which is part of the company’s SQL Server.

1.7 getting heterogeneous Data into the Warehouse

All data warehouses store their data grouped together by subject areas that reflect the general
usage of the data (Customer, Product, Finance etc.). The general principle used in the majority

LoveLy professionaL university 13