Size estimation of olap systems

Jan Zizka (Eds) : CCSIT, SIPP, AISC, PDCTA - 2013
pp. 431–441, 2013. © CS & IT-CSCP 2013 DOI : 10.5121/csit.2013.3649
SIZE ESTIMATION OF OLAP SYSTEMS
Madhu Bhan1
, T V Suresh Kumar2
and K.Rajanikanth3
1,2
M.S.Ramaiah Institute of Technology-Bangalore,India
3
Visvesvaraya Technological University, India
madhoobhan@yahoo.co.in
ABSTRACT
Software size estimation at early stages of project development holds great significance to meet
the competitive demands of software industry. Software size represents one of the most
interesting internal attributes which has been used in several effort/cost models as a predictor
of effort and cost needed to design and implement the software. The whole world is focusing
towards object oriented paradigm thus it is essential to use an accurate methodology for
measuring the size of object oriented projects. The class point approach is used to quantify
classes which are the logical building blocks in object oriented paradigm. In this paper, we
propose a class point based approach for software size estimation of On-Line Analytical
Processing (OLAP) systems. OLAP is an approach to swiftly answer decision support queries
based on multidimensional view of data. Materialized views can significantly reduce the
execution time for decision support queries. We perform a case study based on the TPC-H
benchmark which is a representative of OLAP System. We have used a Greedy based approach
to determine a good set of views to be materialized. After finding the number of views, the class
point approach is used to estimate the size of an OLAP System The results of our approach are
validated.
KEYWORDS
Class point, Materialized views, OLAP Systems, Software size, performance
1. INTRODUCTION
In the present day scenario, estimating the size of the software has become a tedious task. Size
evaluation is one of the main tasks for planning software project development with reliable cost,
effort and performance estimation [15]. The applicability of Function Point (FP) approach to
estimate software size is limited to procedure oriented systems [11]. FP is not suitable for object
oriented programming paradigm which involves classes, encapsulation, inheritance and message
passing [3], [5]. The idea underlying the Class Point method is the quantification of classes in a
program in analogy to the function counting performed by the Function Point measure. In the
procedural paradigm the basic programming units are functions or procedures; whereas, in the
object-oriented paradigm, the logical building blocks are classes, which correspond to real-world
objects and are related to each other [1],[8]. In this paper, we explore the Class Point approach for
estimating the size of OLAP systems based on design documentation. Fast analysis of data stored
in databases or warehouses, is indispensable for businesses that wish to stay ahead in the present
competitive market scene. A data warehouse is a relational database that usually contains
historical data derived from multiple, heterogeneous and independent data sources [16].
Materialized views are views that improve query execution times by pre-calculating expensive

432 Computer Science & Information Technology (CS & IT)
joins and aggregation operations prior to execution of queries and storing the results at the data
warehouse [7],[8]. This dramatically improves the response time of decision support queries.
However the number of possible views exponentially increases relative to the number of database
dimensions. We need to find out what views should be materialized in order to improve query
performance under resource constraints. A greedy algorithm is adopted to choose the most
beneficial view per storage space (Benefit-Per-Unit-Space) up to the given storage limit. The
algorithm considers that there is a linear relationship between the cost of answering a user query
and the size of the view that is used to answer that query. This cost, which is the number of rows
in the view, is then used to select the most beneficial view for materialization. In this paper we
study the TPCH Benchmark [6] to find the number of tables and views present in an OLAP
system.
2. CLASS POINT APPROACH
The Class Point approach provides a system-level estimation of the size of OO products. The
class point approach was introduced in 1998 [17]. In object-oriented development, the class
diagram has a great deal of quantification information based on the design document. It contains
the structural functionality of the target system and its class hierarchy, which are the logical
blocks of the developed system. This approach to size estimation focuses on 1) Local methods
2)Interaction of the class 3) The attributes. The Class Point size estimation process is structured
into three main phases, corresponding to analogous phases in the FP approach. During the first
step the design specifications are analyzed in order to identify and classify the classes into four
types of system components, namely the problem domain type, the Human interaction type, the
data management type, and the task management type. During the second step, each identified
class is assigned a complexity level, which is determined on the basis of the local methods in the
class and of the interaction of the class with the rest of the system. The measures and the way
they are used to carry out this step represent the substantial difference between CP1 and CP2.
Indeed, in CP1 the complexity level of each class is determined on the basis of the Number of
External Methods (NEM) and the Number of Services Requested (NSR). The NEM measure of a
class in an object-oriented system is given by the number of public methods in a class. NSR is
used to measure the interconnection of system components. It is again applicable to a single class
and is determined by the number of different services requested to other classes. In CP2, besides
the above measures, the Number Of Attributes (NOA) measure is also taken into account in order
to calculate the complexity level of each class. Once a complexity level of each class has been
assigned, such information and class type are used to assign a weight to the class. Then, the Total
Unadjusted Class Point value (TUCP) is computed as a weighted sum. The Technical Complexity
Factor (TCF) of the application is determined by assigning the degree of influence that 18
general system characteristics have on the application. The sum of the influence degrees related
to such general system characteristics forms the Total Degree of Influence (TDI). Finally, the
Class Point value is determined by adjusting the TUCP with a value obtained by considering
global system characteristics as in FPA and some additional characteristics especially conceived
for object-oriented systems, namely:
• User Adaptivity
• Rapid Prototyping
• Multiuser Interactivity
• Multiple Interfaces.
3. RELATED WORK
Many metrics for size estimation have been proposed for procedure oriented systems among
which the Function points have achieved a wide acceptance in the estimation of size of business

Computer Science & Information Technology (CS & IT) 433
systems [5]. The method provides estimation of size by measuring the functionality of the
system to be developed.. This allows Function Point Analysis (FPA) to be applied in the early
phases of the lifecycle, which is the main reason for the success of the method. Function point
depends on the information available at the time of specifications [11].Several measures have
been defined so far in order to estimate the size of software systems. Chidamber and Kemerer
defined six measures for assessing OO systems.[12] Among them the Weighted Method per
Class(WMC) , the Number of Children and the Response for a Class have been used as size
measures. All of these measures are useful for productivity analysis as all of them provide class
level measurement. The Use Case Points approach was introduced by Karner [4] as a software
project effort estimation model. Use Case Point effort estimation is an extension of existing
estimation methods, such as function point analysis. This approach, however, has weak points
when applied to general software projects. In recent past other size measures have been
suggested as adaptations of the FP method to Object oriented systems [2]. Whitmire proposes the
application of his 3D Function Points to object-oriented software systems, by considering each
class as an internal file and messages sent across the system boundary as transactions [14].
However, 3D Function Points require a greater degree of detail in order to determine size and
consequently make early counting more difficult. Object point measure is another adaption of
function point, used in the improved COCOMO2.0 effort estimation technique[13]. Object Point
count is very similar to Function Point, but objects are taken as the basis of the counting process.
However, such objects are not directly related to objects in the OO methodology, but rather refer
to screens, reports, or 3GL modules. The class point approach is conceived by recasting the ideas
underlying the FP analysis within the OO paradigm and by combining well known OO measures
[1],[3],[9]. The class point size estimation process is structured in three main phases and two
levels of complexity and classifying the system into component types.
4. OLAP SYSTEM
OLAP systems have become increasingly popular in many application areas as they considerably
ease the process of analysing large amounts of data, stored in data warehouses [16]. Data
warehouse must have efficient OLAP tools to explore the data and to provide users with real
insight of the data in data warehouse. Due to large size of the data warehouse and the complexity
of queries, quick response time plays an important role as timely access to information is the
basic requirement of an OLAP system. Unified Modelling Language (UML) diagrams represents
the static and dynamic aspects of a OLAP system. The UML class diagram in Figure.1 shows the
static structural behaviour of the OLAP system, in which operations are designed for the complete
system. The class diagram has persistent classes, like Dimensions ,Facts and Views and Control
classes like ORB, Query Execution, OLAP API, OLAP operations, and Aggregation. These
classes are related to each other by through associations. The access layer must be able to
translate the data related requests from the user or business layer i.e it must be able to create the
correct SQL statement and execute it. Server programs generally receive requests from the client
from the client programs and execute database retrieval and updates. Each portion of the database
is managed by a server, a process which is responsible for controlling access and retrieval of data
from database portion. The server dispenses information to client applications. The client and the
server processes communicate through a well defined set of standard application program
interfaces(API's).The data model incorporated into a database system defines a framework of
concepts that can be used to express the problem domain. Materialized views within the data
warehouse are transparent to the end user or to the database application. Materialized views are
usually accessed through the query rewrite mechanism. If a materialized view is to be used by
query rewrite, it must be stored in the same database as the fact or detail tables on which it relies.
The motivation for using materialized views is to improve performance. Materialized view
management activities considers measuring the space to be used by materialized views
,determining which existing materialized views should be dropped, ensuring that all materialized

views are refreshed properly each time the database is updated. In such a system, aggregates play
a very important role, because an OLAP query is usually an aggregated view on existing
(relational) data. In SQL, you are probably familiar with aggregate functions like: COUNT, SUM,
AVG, MIN and MAX. Execution of query by the relational engine involves parsing of the
submitted statements, optimization of the SQL statements, compilation of the code, and
generation of the query execution plan. During execution, programs call the storage engine to
retrieve or manipulate the data stored in the database. A database administrator adds OLAP
metadata to a data warehouse. The end result is the creation of one or more measure folders that
contain one or more measures. The measures have dimensions, and the dimensions have
hierarchies, levels, and attributes. An OLAP API gives access only to the measures that are
contained in measure folders. Conceiving data as a cube with hierarchical dimensions leads to
conceptually straight forward operations to facilitate analysis. Common OLAP operations include
roll up, slice and dice, drill down and pivot. A roll-up involves summarizing the data along a
dimension. Drill Down allows the user to move from the current data cube to a more detailed data
cube. Slice is the act of picking a rectangular subset of a cube by choosing a single value for one
of its dimensions, creating a new cube with one fewer dimension. The dice operation produces a
subcube by allowing the analyst to pick specific values of multiple dimensions. The Object
Request Broker (ORB) is a process which sends and receives messages to resources and other
services distributed across multiple application servers. CORBA object request brokers(ORB's)
implement a communication channel through which applications can access object interfaces and
request data and services. CORBA requires an Object Request Broker both on the OLAP API
client computer and on the OLAP Services computer. When an application calls a method that
requires an interaction with an OLAP server, the client ORB intercepts the call, interacts with the
OLAP servers ORB to find the object on the server side that can implement the request, passes
the parameters, invokes the object's method, and returns the results. The client application forms
the front end of the system which the user sees and interacts with. The end-user query model
identifies all the conceptual query objects with which the application user interface will deal, thus
taking advantage of the strengths of object-oriented design. It allows for a clear correspondence
between user interface objects and OLAP API objects. The UML class diagram in Figure.1 shows
the static structural behaviour of the OLAP system, in which operations are designed for the
complete system. The class diagram has persistent classes, like Dimensions ,Facts and Views and
Control classes like ORB, Query Execution, OLAP API, OLAP operations, and Aggregation.
These classes are related to each other by through associations. The steps to calculate the class
point for OLAP Systems are as follows:
1.Identification of classes
2.Determination of complexity of classes
3.Calculation of unadjusted class point
4.Calculation of technical complexity factor
5.Calculation of class point

Figure 1.Class diagram
Step 1.
From the class diagram of OLAP system, the classes are classified into typical PDT, HIT, DMT
and TMT classes as given in Table 1.
Step 2.
The class point method uses two complexity level measures CP1 and CP2. In CP1 the complexity
level of each class is determined based on Number of external methods and Number of services
requested. Both the measures are available in design documentation. The CP1 measure can be
used in early phases of software development and the CP2 measure can be used only when the
number of attributes is available [1]. Thus considering only CP1 the complexity and the weights
associated with various classes forming an OLAP system is given in Table 2.

Table 1
Table 2
Step 3.
Assuming that there are n dimensions and facts and m number of materialized views we can
compute total unadjusted class point value. Thus, the TUCP is computed as the weighted total of
the four components of the application:
4 3
TUCP =Σ Σ w ij × xij;
i=1 j=1
where xij is the number of classes of component type i (problem domain, human interaction, etc.)
with complexity level j (low, average, or high), and wij is the weighting value for type i and
complexity level j.
Class type Description OLAP Systems
Prolem Domain
Type(PDT)
Represents real world entities in the
application domain
Dimensions and Facts
Human Interface
Type(HIT)
Satisfies the need for visualizing
information and human computer
interactions.
User-Interface
Data Management
Type(DMT)
Includes classes which Incorporate data
storage and retrieval
OLAP server, DW server,
MVS, Metadata, Mat. views
Task Management
Type(TMT)
Includes classes which are responsible
for tasks
ORB, Query Exec, OLAP API,
OLAP operations, Aggregation
Class Type NEM NSR Complexity Weight.
Fact, Dimension PDT 3 0 low 3
MVS DMT 0 2 Avg. 8
OLAP server DMT 2 10 Avg. 8
DW Server DMT 1 7 High 13
Metadata DMT 3 2 Low 5
Aggregate DMT 5 2 Avg. 8
Mat. Views DMT 3 0 Low 5
ORB TMT 3 0 Low 4
OLAP API TMT 4 8 Avg. 6
Query Exec TMT 4 4 Avg. 6
OLAP Operation TMT 5 5 High 9
User Interface HIT 0 5 Avg. 7

Step 4.
The Technical Complexity Factor (TCF) is determined by assigning the degree of influence
(ranging from 0 to 5) that 18 general system characteristics have on the application from the
designer’s point of view [1]. The sum of the influence degrees related to such general system
characteristics forms the Total Degree of Influence (TDI), which is used to determine the TCF
according to the following formula:
18
TCF = 0.55 + 0.01× Σ f i
i=1
For OLAP Systems characteristics like Data communication, Distributed functions, Multiple
users, Ease of operation ,adaptability by user have strong or significant influence on development
of the system while as characteristics like Transaction rate, Online data entry, Online update and
multiuser interaction have no influence or least influence on processing the system. The other
characteristics Performance, Reusability, Compiler processing and High end configuration have
average influence on the system. Based on the influence of these characteristics the TCF
factor is calculated as
TCF=0.55+0.01{55}=0.55+0.55=1.10
Step 5.
The final value of the Adjusted Class Point (CP) is obtained by multiplying the Total Unadjusted
Class Point value by TCF .
CP = TUCP×TCF
The CP count can vary with respect to the unadjusted count from -45 percent +45 percent due to
adjustment factor. For OLAP systems the final class point value is given as
CP=1.10(74+3n+5m) --- eq. 1.
Where n is the number of facts and dimensions and m is the number of materialized views.
Experimental Study
5. EXPERIMENTAL STUDY
In order to determine the values of m and n in the above Complexity formula we have used TPC–
H Benchmark for experimentation and Illustration [6]. The TPC Benchmark H models a data
warehouse for any organization which must sell, distribute or manage a product worldwide. The
data base has data about each such transaction over a period of seven years. The TPC-H database
is defined to consist of eight separate tables. The name of the tables in itself indicate their
contents: part, supplier, partsupp, customer, nation, region, lineitem and orders The queries and
the data populating the database have been widely used in research as it has industry-wide
relevance. Using Table-Class mapping where a single table is mapped to a single class we have
obtained the value of n as 8 [2]. In order to obtain the value of m i.e the number of materialized
views we need to understand the Lattice structure of a cube and the concept behind Greedy
algorithm for selecting the views. The challenge for the design of OLAP systems is the
exponential explosion of possible views as the number of dimensions increases. If D is the

number of dimensions and hi the number of hierarchical levels in dimension i then the general
equation for calculating the number of possible views is given by Equation
D
Possible views = Π hi
i=1
As dimensionality increases linearly, the number of possible views explodes exponentially.
OLAP system cannot materialize all the views in a given lattice structure because of constraints
on storage space, computational time and view maintenance cost.. Typically, a strategic subset
of views must be selected for materialization. There is a need to select an optimal set of views to
be materialized. Using Greedy based approach for view selection we select a beneficial view at
each step that fits within the space available for view materialization. Greedy algorithm
considers Cost(vw), the cost associated with each view based on the number of rows in the
view. k is the number of views to be materialized in addition to the root view. After selecting
set S of views, the benefit of view vw relative to S is denoted by Ben(vw,S). For selecting a set
of k views to be materialized ,the Greedy Algorithm is given below:
S = {root view};
for i = 1 to k do begin
select that view vw not in S such that Ben(vw,S) is maximized;
S = S union {vw};
end;
For our experimental study, we have populated a 1-GB Benchmark database. We populated the
root node from this database using the Customer (C), Part (P), Month (M) dimensions, and the
“Sale” measure. The customer dimension has 3 levels of hierarchy customer, nation and all. The
part dimension has 3 levels of hierarchy part, part type, all. The sales are analyzed at three levels
of time hierarchy-Month, Year, All. We ran the greedy algorithm on the lattice of Figure 2 using
the TPC-H database as described above. When k=5, the greedy algorithm picks the root view
(view 0) view then view 6 whose benefit is maximum as all the views below it are each improved
from 6001192 to 45000,when we use view 6 in place of view 0. Similarly view 9,view 10,view 11
are picked according to the order of their respective benefits. Table 3 shows the results for the
number of views to be materialized derived after applying Greedy Algorithm. In the first problem
instance, we imposed a constraint of 5 views to be materialized; in the second problem instance,
we set it to 10 views; in the third.

.
Figure 2:Lattice Structure with Actual Number of Rows
(Source:TPCH Benchmark Database)
Table 3
Lattice Problem Instance Optimal solution
(Set of Materialized Views)
TPC-H
Materialize 5 views 0,6,9,10,11
Materialize 10 views 0,2,6,9,10,11,12,15,20,21
Materialize 20 views 0,2,3,4,5,6,7,8,
9,10,11,12,13,14,15,16,17,18,20,21,24

Figure 3
problem instance, we set it to 20 views. The views picked up by the greedy algorithm with
maximum benefit in terms of storage space are shown in the table. The Y-axis in the graph of
Figure 3 shows the total time taken as well as the space used. On X-axis it has the number of
views picked. From the graph we can make a clear decision of when to stop materializing views.
It is clear from the graph that for the first 10 views the query time in terms of number of rows is
reduced substantially. However we have observed that performance after materializing 10 views
remains constant and hence we do not materialize the remaining possible views. After knowing
the number of tables and number of views we can calculate the value of CP1 as given in eq. 1
which equals to 158. The Effort is defined using regression analysis [1] as
Effort=0.843*CP1+241.85. Thus the effort comes to be equal to 380 person hours.
6. VALIDATION AND CONCLUSION
Size estimation is one of the critical tasks in object oriented software project management. It is
widely accepted that system size is strongly correlated with development effort [13], [18], [19],
[20]. The Class Point approach provides a system-level size measure by suitably combining well
known OO measures, which consider specific aspects of a single class. In particular, two
measures are proposed, namely, CP1 and CP2. We have used CP1 at the beginning of the
development process of OLAP System to carry out a preliminary size estimation. In this paper we
have used UML class diagram document for size estimation of OLAP systems. The results
of our approach are validated from websites like - http://datawarehouse.ittoolbox.com, which
estimate effort of OLAP Systems based on their complexity to be as 1) Simple OLAP - 2
weeks 2) Medium OLAP - 2 months 3) Complex OLAP - 2 years. Assuming Medium case
OLAP system where the data warehouse is already existing, our approach to size estimation is
very close to industry estimated values. Future work may include refinement of size estimation
by applying CP2.
REFERENCES
[1] Gennaro Costagliola and Genoveffa Tortora, “Class Point: An Approach for the Size Estimation of
Object-Oriented Systems”, IEEE transaction on Software Engineering, Vol. 31, No. 1,January
2005,page 52-74.
[2] Ali Bahrami, Object-Oriented System Development , International edition , 1999.

[3] Wei Zhou and Qiang Liu,” Extended Class Point Approach of size Estimation for OO Product”,
IEEE sponsored 2nd International Conference on Computer Engineering and Technology,2010,Vol-
Page:117-122.
[4] Parastoo Mohagheghi, Bente Anda, and Reidar Conradi, “Effort Estimation of Use Cases for
Incremental large-Scale Software Development, ” Proceedings of the International Conference on
Software Engineering (ICSE’05), pp. 303-311, May 15-21. 2005.
[5] Roger S. Pressman, ―Software Engineering A practioner’s Approach, published McGraw Hill,
2005.
[6] TPC Benchmark H (Decision Support) Revision 2.14.4. http://www.tpc.org.
[7] Yang J., Karlapalem K., Li Q., Algorithms For Materialized View Design in Data Warehousing
Environment. VLDB 1997.
[8] Andreas Rauber, Philipp Tomsieh, ”An Arichitecture for modular On- Line Analytical Processing
systems:Supporting Distributed and Parallel Query Procesing using Co-operating CORBA objects”.
DEXA Workshop,1999.pp45-49.
[9] Aditi Kapoor, Parul Pandey ,”Fuzzy Class Point Approach”.Journal, IJRTET,Volume 5, Issue 1,
2011.
[10] V. Harinarayan, A. Rajaraman, and J. D. IJllman. Implementing Data Cubes Efficiently. A f~dl
version of this paper. At http://clb.stanford. edu/ pllb/harinarayan /199,5/cube.ps
[11] J.B. Dreger, Function Point Analysis. Prentice Hall, 1989.
[12] C.F. Kemerer and B.S. Porter, “Improving the Reliability of Function Point Measurement: An
Empirical Study,” IEEE Trans. Software Eng., vol. 18, no. 11, pp. 1011-1024, Nov. 1992.
[13] P. Nesi and T. Querci, “Effort Estimation and Prediction of Object- Oriented Systems,” J. Systems
and Software, vol. 42, pp. 89-102, 1998.
[14] S.A. Whitmire, “3D Function Points: Applications for Object- Oriented Software,” Proc. ASM’96
Conf., 1996.
[15] L.C. Briand, S. Morasca, and V.R. Basili, ―Property Based Software Engineering Measurementǁ,
IEEE Transaction on Software Engineering, vol. 22, no. 1, pp. 68-86, 2009.
[16] Surajit Chaudhuri,Umeshwar Dayal,”An Overview of Data Warehousing and OLAP Technology”
.SIGMOD Record, Vol.26,No.1,1997.pp 65-74 .
[17] Smith R. K, “Effort Estimation in Component-Based Software Development: Identifying
Parameters”, In the proceedings of the 36th ACM Southeast Regional Conference, Marietta, Georgia,
USA, pp 323- 325,1998.
[18] V.B. Misic and D.N. Tesic, “Estimation of Effort and Complexity: an Object Oriented Case Study,”
J. Systems and Software, vol. 41, pp. 133-143, 1999.
[19] S. Moser, B. Henderson-Sellers, and V.B. Misic, “Cost Estimation Based on Business Models,” J.
Systems and Software, vol. 49, pp. 33- 42, 1999.
[20] B.W. Boehm, B. Clark, and E. Hiriwitz, “Cost Models for Future Life Cycle Processes: COCOMO 2.
0,” Ann. Software Eng., vol. 1, no. 1, pp. 1-24, 1995.
Authors
Madhu Bhan. Faculty at M.S.Ramaiah Institute of Technology, Bangalore, India.
Currently perusing P.hd. Area of Intrest is Databases and Data Warehousing
Dr.T.V. Suresh Kumar. Professor at M.S. Ramaiah Institute of Technology,
Bangalore, India. Area of Interests include Software Performance Engineering,Object
Technology.
Dr.K.Rajanikanth. Ph.d from Indian Institute of Science (IISc), Bangalore, India.
Currently working as Chairman and Advisor for many bodies of Visvesvaraya Tech.
University. Area of interests include Adhoc Networks, Image processing,
Microprocessors and Micro-controllers, Data Warehouses.

Size estimation of olap systems

More Related Content

What's hot (18)

Viewers also liked (15)

Similar to Size estimation of olap systems (20)

Recently uploaded (20)

Size estimation of olap systems