OLAP, MOLAP, ROLAP 1

[ADB1]

Could someone please provide a definition of the above in their simplist form.......

Thanks,

Adam.

 

[Patten]

There are two prominent architectures for OLAP systems: multidimensional OLAP
(MD-OLAP) and relational OLAP (ROLAP). MD-OLAP architectures utilize a multidimensional
database to provide analyses; their main premise is that OLAP is best
implemented by storing data multidimensionally. In contrast, ROLAP architectures
access data directly from data warehouses; ROLAP architects believe that OLAP capabilities
are best provided directly against the relational database.When comparing
these two architectures, the following observations can be made:
• ROLAP leaves the design trade-off between query response time and batch processing
requirements to the system designer, as the ROLAP architecture is neutral to the amount
of aggregation in the database. MD-OLAP generally requires most of the database to
be precompiled in order to provide acceptable query performance, thereby increasing
batch processing requirements.
• Systems with high data volatility, namely those with changing data aggregation rules
and user-defined consolidations, require an architecture that can dynamically consolidate
data for ad hoc and decision support analyses. ROLAP is very well suited for
dynamic consolidations whereas MD-OLAP is biased towards batch consolidations.
• ROLAP can scale to a large number of business analysis perspectives (dimensions),
while MD-OLAP generally performs efficiently with ten or fewer dimensions.
• ROLAP supports OLAP analyses against large volumes of input (atomic-level) data.
In contrast, MD-OLAP provides adequate performance only when the input data set
is small (fewer than five gigabytes).
Conclusion:ROLAP is a flexible, general architecture that scales to meet the widest
variety of DSS and OLAP needs. MD-OLAP is a particular solution that is suitable
for departmental systems with small data volumes and limited dimensionality.

 

[jnicho02]

The base query involved in OLAP is the 'rollup'. In SQL this is achieved by a 'group by' statement.

If you have 3 dimensions: age, sex and marital status you can answer loads of questions with the result set from:

select age, sex, marital_status, count(*)
from fact
group by age, sex, marital_status

for example, with a result set of.....
age, sex, marital_status, count(*)
23 , M, single, 12
23 , M, married, 2
23 , M, divorced, 0
23 , F, single, 8
23 , F, married, 3
23 , F, divorced, 1

Q. how many males? 12+2 = 14
Q. how many married people? 2+3 = 5
Q. how many married males? 2
Q. %age males who are married? 2/14 * 100 = 14%

The point is, as we have a summary we have to look at fewer rows of data to get the answers.

The basic difference between ROLAP and MOLAP is that in Relational OLAP we will ask the database to do the group-by at the time and then work with the result, and in MOLAP we will pre-rollup the whole database.

ROLAP is flexible and quick as long as the amount of data it has to roll-up is limited. MOLAP can lead to very quick results for users as it is precompiled. Jeremy Nicholson, Director of

http://www.firstlightassociates.co.uk

a UK-based Java and Data Warehousing consultancy

 

[jnicho02]

...and, as Patten says, ROLAPs can be tuned by having strategically pre-compiled aggregate tables to gain the best of both worlds. Jeremy Nicholson, Director of

http://www.firstlightassociates.co.uk

a UK-based Java and Data Warehousing consultancy