Analysis of Data Warehouse Architectures: Modeling and Classiﬁcation

Qishan Yang

, Mouzhi Ge

and Markus Helfert

Insight Centre for Data Analytics, Dublin City University, Dublin, Ireland

Faculty of Informatics, Masaryk University, Brno, Czech Republic

Keywords:

Data Warehouse, Architecture, Classiﬁcation, Modeling, Big Data, Archimate.

Abstract:

With decades of development and innovation, data warehouses and their architectures have been extended to a

variety of derivatives in various environments to achieve different organisations’ requirements. Although there

are some ad-hoc studies on data warehouse architecture (DWHA) investigations and classiﬁcations, limited

research is relevant to systematically model and classify DWHAs. Especially in the big data era, data is

generated explosively. More emerging architectures and technologies are leveraged to manipulate and manage

big data in this domain. It is therefore valuable to revisit and investigate DWHAs with new innovations. In this

paper, we collect 116 publications and model 73 disparate DWHAs using Archimate, then 9 representative

DWHAs are identiﬁed and summarised into a ”big picture”. Furthermore, it proposes a new classiﬁcation

model sticking to state-of-the-art DWHAs. This model can guide researchers and practitioners to identify,

analyse and compare differences and trends of DWHAs from componental and architectural perspectives.

1 INTRODUCTION

In an organisation, data usually come from heteroge-

neous resources that include different views and may

provide inconsistent information to end users. Thus,

an organisation needs a systematic channel to main-

tain and manipulate its data. Although operational

database management systems (ODBMS) could par-

tially solve the problem, it mainly serves for running

daily business transactions, which focuses more on

managing active interactions and data on time. There-

fore, a Data Warehouse (DWH) is necessary to man-

age the business and support strategic decisions for

an organisation (Ge and Helfert, 2006). (Devlin and

Cote, 1996) states that a DWH is a single, complete,

and consistent store of data from various sources and

made available to end users in a business context.

This centralised data system governs and maintains

data for analyses to meet an organisation’s require-

ments.

DWHs need to rely on architectures and other

components assisting them, e.g. solving data qual-

ity issues and integrating data (Yang et al., 2017).

Data warehouse architectures (DWHAs) are devel-

oped synchronously along with DWHs’ progress to

mainly serve for DWHs. A variety of DWHAs is pro-

posed in order to ﬁt different requirements and situ-

ations. To the best of knowledge, there is no clas-

siﬁcation which categories all the DWHAs. (Ariy-

achandra and Watson, 2008) and (Matouk and Owoc,

2012) conduct surveys on parts of DWHAs and cal-

culate their distributions mainly in the United States

and Poland respectively. They only cover ﬁve types

of DWHAs. (Hub-and-Spoke, Data Mart Bus, Cen-

tralised, Independent Data Marts and Federated). The

latest DWHA classiﬁcation is proposed by (Bla

et al., 2017), but some DWHAs, such as Virtual

DWHA, Big DWHA and DWH with data lake ar-

chitecture (DLA), are still excluded. These excluded

DWHAs also frequently appear in literature and in-

dustry, which solve speciﬁc problems in terms of ad-

dressing big data issues (Inmon, 2016) and swift data

analyses (Pasquale et al., 2008). While, the DHWAs

included in their research above have few advantages

to solve these problems, which would be more time-

consuming and less-efﬁcient. Hence, it is worthy to

discuss and classify them together. It could beneﬁt

people having more choices to achieve their needs es-

pecially in big data and high-efﬁcient era.

In order to systematically conduct this research,

a systematical literature review is taken to collect

DWHAs in literature. Due to different descriptions

of DHWAs, they are generalised and modelled in or-

der to better demonstrate and differentiate them. It

further analyses the distribution of each DWHA in

literature then compares the result with two previous

604

Yang, Q., Ge, M. and Helfert, M.

Analysis of Data Warehouse Architectures: Modeling and Classiﬁcation.

DOI: 10.5220/0007728006040611

In Proceedings of the 21st International Conference on Enterprise Information Systems (ICEIS 2019), pages 604-611

ISBN: 978-989-758-372-8

surveys to give insights from industrial and academic

perspectives. These DWHAs identiﬁed from litera-

ture are classiﬁed for better structuring and presenting

their relationships. In addition, as intrinsic features of

DWHAs, some of them are analogous and difﬁcult

to distinguish. Hence, it is necessary to discuss and

compare their dissimilitudes. The main contributions

of this research are summarised below.

• identiﬁed a plethora of DWHAs based on a sys-

tematic literature review;

• modelled and generalised DWHAs to describe

similarities and differences among them;

• conducted an analysis of DWHAs to offer an

overview of their distributions;

• proposed a classiﬁcation model to analyse differ-

ent types of DWHAs.

The remainder of the paper is organised as fol-

lows. Section 2 reviews the related work of previ-

ous research in this domain. Section 3 presents the

research methods applied in this research. Section

4 explains the results based on data from literature.

Section 5 proposes a DWHA classiﬁcation model and

further analysis. Finally, section 6 summarises this

research and outlines future work.

2 RELATED WORK

With decades of development, DWHs and their archi-

tectures are not only focusing on data analyses and

decision-making, but they also follow closely behind

the ever-accelerated updating of science and technol-

ogy to meet new requirements. This paper mainly

focuses on investigating DWHAs. As the foremost

components in these architectures, DWHs also should

be concerned and discussed hereby. There are dif-

ferent types of traditional DWHAs to address size-

controlled data sets. They are extended to different

types (e.g. big DWH and data lake) in terms of the big

data challenge. Some relevant concepts are discussed

in this section to clear the boundary of this research.

2.1 Traditional Data Warehouse

A traditional DHW herein normally refers to a

DWH mainly addressing size-controlled structured

data (Katal et al., 2013). Different understandings of

DWHAs can be found in literature. (Inmon, 1997)

deﬁnes that a DWH is a subject-oriented, integrated,

non-volatile and time-variant collection of data that

supports of management’s decision-making capabili-

ties. (Kimball and Ross, 2013) advocates that a DWH

is the conglomerate of all DMs within the company

and information is always stored in the dimensional

model. Two main approaches of developing DWHs

are the top-down and bottom-up (George, 2012). In

the top-down approach, a centralised DWH is de-

veloped ﬁrst, then building independent data marts

(DMs) for individual departments. However, in the

bottom-up approach, DMs are built ﬁrst then the

DWH is created by combining these DMs.

In addition, there is another terminology called

”data warehousing”, which is similar to the DWH but

with a different meaning. Data warehousing more

focuses on processes or mechanisms of how to or-

ganise algorithms, components and data to build data

warehouses. When comparing the DWH and data

warehousing, the former more focuses on storing and

managing data with several characteristics mentioned

above, while the data warehousing lays emphasis on

the process of building a DWH project. Hence, when

collecting data, this terminology is not included in this

research.

2.2 Big Data Warehouse

Nowadays, DWHs are facing exploded data that is

usually referred to as big data. Especially, unstruc-

tured data dominates a large portion of big data.

This type of data contributes and drives the innova-

tion which should be concerned if more data needs

to be involved in the process of analysis. It consti-

tutes 90 percent of information, which includes so-

cial text, audio, video, sensors, and click stream data,

etc. While, the traditional DWH systems are not de-

signed to scale, and handle this exponential growth of

unstructured data (Pasupuleti and Purra, 2015). In or-

der to address new unstructured big data challenges,

DWHs are extended and merged with state-of-the-

art technologies. The big data platforms or concepts

(e.g. Hadoop and Hive) are leveraged to build DWHs

with high-speed processing capacities (Thusoo et al.,

2010b).

These DWHAs associated with state-of-the-art

technologies are indispensable to address big data is-

sues especially in manipulating a mass of unstruc-

tured data. For instance, (Thusoo et al., 2010b) ap-

plies Hive on top of Hadoop to develop a big DWH to

store and analyse log data at Facebook. This research

mainly focuses on the reconciled layer. (Yang and

Helfert, 2016b) investigates and builds a data ware-

house architecture in the context of big data. It is

based on the Flume, Hadoop, Hive, HBase, Sqoop etc.

to establish a DWHA to store and analyse website log

data. In this research, the reconciled layer and derived

layer are built and mixed in the Hadoop platform.

Analysis of Data Warehouse Architectures: Modeling and Classiﬁcation

605

2.3 Data Lake

In addition, as a novel data repository, the Data Lake

(DL) is also an option to address big data issues. Ac-

cording to the literature, a DL not only can be applied

to manage data solely, but it can also be accompa-

nied by a DWH. (Research, 2014) holds that the DL

and the enterprise DWH provide a synergy of capabil-

ities to deliver accelerating responses. They cooperate

with each other to present their competences respec-

tively. This combination enables people to deal with

enormous data and driving business results. There are

different descriptions and deﬁnitions of the DL and its

architecture. James Dixon proposed that data in a DL

is more natural than the data in a DM in which data is

cleansed, packaged and structured for easy consump-

tion (Dixon, 2010). A DL is a collection of vari-

ous data assets which are near-exact or even exact of

the source format. It serves only for the most highly

skilled analysts to explore data.

A DL is thus less structured compared to a tra-

ditional and big DWHs as data preparation is elim-

inated. It is introduced to maintain and manipulate

the explosively growing unstructured data. This novel

system has abilities to capture and maintain various

types of raw data with a low cost, operate transforma-

tions on the data, deﬁne the schema on the ﬂy, per-

form new types of data processing, answer ad hoc

analyses based on the speciﬁc use cases (Research,

2014). (Devlin, 2014) summarises that a DL is an

idea that all enterprise data can and should be stored in

Hadoop. It is a rip-and-replace strategy for all DWHs,

DMs and eventually even operational databases be-

cause of its fullest extent.

2.4 DWHA Classiﬁcations

According to the literature, there are two main

DWHA classiﬁcations which observe DWHAs from

different views. The ﬁrst classiﬁcation includes

single-layer, two-layer and three-layer architectures

that are depending on the number of layers used by

the architecture (Devlin and Cote, 1996; Senapati

and Kumar, 2014). The second classiﬁcation con-

sists of Independent DMs architecture, Bus, Hub-and-

Spoke, Centralized and Federated architectures based

on components and relationships (Ariyachandra and

Watson, 2008).

In terms of these three types of DWHAs, The pri-

mary mechanism of the single-layer (real-time layer)

is all data sets are stockpiled once only (Inmon, 1997).

An operational system and a data warehousing sys-

tem share the identical data that are stored in the same

place. The two-layer architecture adds a derived data

layer to enable the operational and analytical require-

ments to be fulﬁlled separately. This approach ad-

dresses the single-layer architecture problems, but it

also has the limitation, which has low performance in

inhomogeneous or large volumes of data from differ-

ent sources (Devlin and Cote, 1996). The three-layer

architecture is comprised of a real-time data layer, a

reconciled data layer and a derived data layer, which

adds a reconciled layer as the new layer compared

with the two-layer architecture. This new layer ma-

terialises the integrated and cleansed data.

When investigating DWHAs from components

and their relationships, they can be mainly cate-

gorised into ﬁve predominant architectures (Ariy-

achandra and Watson, 2006; Ariyachandra and Wat-

son, 2010). Although other architectures (e.g. hybrid)

are mentioned in this domain, they tend to be vari-

ations on these ﬁve (Scheibe and Nilakanta, 2008).

(Ariyachandra and Watson, 2008) conducts a Web-

based survey Surveyed companies ranged from small

(less than $10 million in revenue) to large (in excess

of $10 billion). Most of them are located in the United

States (60%) covering a variety of industries. (Ma-

touk and Owoc, 2012) does a similar survey and col-

lected data from ﬁrms listed in the Warsaw Stock Ex-

change. The details of these ﬁve DWHAs and the two

surveys will be discussed in the following sections.

3 RESEARCH METHODOLOGY

In this paper, the quantitative research method is ap-

plied to navigate the processes and output numerical

results. This method is presented as ”quantities” or

numbers, which is normally generate results by do-

ing statistical analyses (Patten and Newhart, 2018). It

normally explains phenomena by analysing collected

numerical data using mathematically based methods

(e.g. statistics)(Aliaga and Gunderson, 2006). There

are different patterns to conduct its processes. (Muijs,

2010) lists four situations of using quantitative meth-

ods. (Muijs, 2010) organises the processes of quanti-

tative research with six steps, which is chosen and ap-

plied in this research: (1) identifying research objec-

tives; (2) selecting research design; (3) deﬁning what

information is needed; (4) executing research instru-

ments; (5) collecting data; (6) analysing data.

Accordingly, our research objectives are set down

in the ﬁrst step, which is investigating the distribu-

tion and classiﬁcation of DWHAs. (Robson and Mc-

Cartan, 2018) proposes three research designs: exper-

imental, quasi-experimental and non–experimental,

which are widely accepted in different domains of

research. In the second step, non–experimental and

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

606

its sub-branch (analysis of existing data sets) are used

for this research. In the third step, the information is

needed in this research including publications regard-

ing DWHAs, hence, the publications are deﬁned with

the keywords of data warehouse architecture. As for

the fourth and ﬁfth, the systematical literature review

is executed as the research instrument to collect data

from existing data sets. The data analyses for the last

step are presented in section 4 and 5.

4 RESULTS

DWHAs are identiﬁed and generalised by analysing

the contents and diagrams of these related publica-

tions collected from literature above. Different de-

scriptions can be found to present even for the same

DWHA in these publications. In order to unify

them, A modelling language, Archimate, is lever-

aged to model and diagram their components and re-

lationships. Archimate is the open source modelling

toolkit for all levels of Enterprise Architects and Mod-

ellers (Modelling, 2018). The conformed modelled

DWHAs can facilitate to observe the similarity and

differentiation of DWHAs.

4.1 DWHA Overview

Archimate is applied to model 73 distinct DWHAs ex-

tracted from 116 articles. Each identiﬁed DWHA is

described and explained below. At the meantime, in

order to better demonstrate and distinguish the differ-

ences among them, these modelled DWHAs are gen-

eralised and merged together in ﬁgure 1.

The Hub-and-Spoke DWHA: Bill Inmon proposes

the Hub-and-Spoke DWHA, because it is suitable for

traditional relational database tools to fulﬁll the de-

velopment requirements of an enterprise-wide DWH

(Breslin, 2004). It is a top-down approach to develop

a DWH system, which means a centralised DWH is

built ﬁrst, then DMs are built on this DWH.

The Data Mart Bus DWHA: Ralph Kimball (Kim-

ball and Ross, 2013) supports the Data Mart Bus us-

ing dimensional modelling, in which a data modelling

approach is unique to the DWH. The Enterprise-

wide cohesion is accomplished by a data bus standard

(Breslin, 2004). It is a bottom-up approach to develop

a DWH. Conformed dimensional tables and DWH bus

matrices play vital roles to provide consistent dimen-

sions for implementation of DMs asynchronously.

The Centralised DHWA: The Centralised DHWA

collects heterogeneous data into the enterprise-level

cross-functional information system, which enables

the separate sources to be used together and stores

Figure 1: Data Warehouse Architecture Overview.

the presentation-ready format for requests from end

users or further analysis. The topology of this DWH is

not complicated to maintain warehouse data for many

end-users and applications throughout the organiza-

tion (Bontempo and Zagelow, 1998).

The Independent DWHA: In this research, the in-

dependent DHWA originally derives from a broader

scope rather than the Independent Data Marts. The In-

dependent Data Marts consist of physically/logically

separated and irrelevant DMs. while the Independent

DHWA may contain hybrid and less-coupling compo-

nents including DWHs and DMs. This type of archi-

tecture may be called as the DM architecture which

has the independent DM(s) or distributed DHWA

that contains the mixed but independent DWH(s) and

DM(s) like the architecture in (Sahama and Croll,

2007).

The Federated DWHA: The Federated DWHA is

based on the “bottom-up” implementation, in which

enterprise-level dependent DMs can provide the

Analysis of Data Warehouse Architectures: Modeling and Classiﬁcation

607

enterprise-wide analytical solutions. The federated

DWH model can be acted as the added data staging

layer that enables easier implementation of analytical

solutions to hide the complexities of operational data

sources (Alsqour et al., 2012).

The Virtual DWHA: The virtual DHW manipulates

and analyses operational data sources with limited

data integration functions and also summary views

of the Data Warehouse depending on the complex-

ity of requirements (Chaki and Sarkar, 2010). Mid-

dleware may be leveraged as an interface between

front-end tools and the operational database (Ayman

et al., 2001). (Devlin and Cote, 1996) describes that

a virtual data warehouse is a logical warehouse al-

lowing users to directly access multiple operational

data through middleware tools. It does not necessar-

ily store copies of the operational data with different

formats compared with other DWHAs.

The Distributed DWHA: The distributed DWHA

herein mainly includes DWHs working in parallel

or built on multi-node cloud computing platforms,

in which data are pre-processed and physically dis-

tributed in ODBMS (e.g Oracle, DB2 etc.) with

a predeﬁned schema. This architecture fragments

the warehouse schema using partitioning algorithms

(Pavlo et al., 2012) and allocates the generated frag-

ments over the compute nodes using particular algo-

rithms(Menon, 2005). In order to gain availability of

data and the high performance of the system, the gen-

erated fragments may be duplicated and saved in dif-

ferent compute nodes.

The Big DWHA: This DWH is built on a big data

platform (e.g. Hadoop), in which the distributed ﬁle

system and other mechanisms (e.g. MapReduce) are

applied to store and deal with data respectively (Yang

and Helfert, 2016b). This DWH can address big data

issues (e.g. petabyte level) for reporting and ad-hoc

analyses by using SQL-like language (e.g Hive) (Thu-

soo et al., 2010a), which is easily executed by peo-

ple with SQL experience. It can manage unstructured

data provided by a logical DWH (Yang and Helfert,

2016a). This logical DWH does not need to strictly

pre-process data with a predeﬁned schema.

The Data Lake Architecture: The DL can be lever-

aged directly as a centralised raw data repository pro-

viding information for further analyses (Rangarajan

et al., 2015). The DL and DWH can also cooperate

together to address the big data issues and achieve

data analysis requirements, which are generalised and

modelled in ﬁgure 1. (Inmon, 2016) presents an ar-

chitecture with a DWH and DL. In their architecture,

the DWH populates the DL, which means the DWH

is built ﬁrst or a company already has a legacy DWH

then build a DL. Another situation is the DWH fed by

the DL (Pasupuleti and Purra, 2015). The DWH ob-

tain data from the DL, which is suitable for building

the DL then the DWH, optimising the legacy DWH

data ﬂows or extending the DWHA to meeting intan-

gible requirements.

4.2 DWHA Distribution Analysis

A statistics is made to analyse the distribution of these

DWHAs in literature and compare with the two pre-

vious research (Ariyachandra and Watson, 2008) and

(Matouk and Owoc, 2012) presented in table 1.

There is no big conﬂict among these three results.

The ﬁrst three DWHAs (Hub-and-Spoke, Data Mart

Bus and Centralised) are similar. In the ﬁrst two re-

search, the ratios of Data Mart Bus are greater than the

Centralised DHWA’s, while in this paper, the result

is the opposite. The reason might be observing the

domain from different angles (e.g. industrial or aca-

demic). Therefore, the three high-weighed DWHAs

are widely accepted in practice and academia. When

choosing or building DWHAs, these three architec-

tures should be concerned. In addition, most of the

papers about DLAs have been published since 2015,

so it shares less proportion. If putting the distributed

DWHA, Big DWHA and DLA together, these three

architectures’ occupation cannot be neglected, which

may imply that big data issues are increasingly in-

escapable in the DWHA context. Hence, when de-

signing and developing DWHAs, it is necessary to

consider these three big-data-oriented architectures.

5 DWHA CLASSIFICATION

MODEL

As discussed before, based on literature, there are two

main DWHA classiﬁcations. The layer-based classi-

ﬁcation is general and abstract, which gives an out-

line of architectures. The component-based classiﬁ-

cation is derived from use cases in practice. However,

they are some inter-relations between them. Some

DWHAs (e.g. Hub-and-Spoke, Data Mart Bus and

Centralised) may need an ETL tool to extract, trans-

form and load data. Some DHWAs (e.g. Federated

data warehouse architecture) may do not need ETL

tools/staging areas, while, in order to build the log-

ical/physical Federated DHW, the data in its under-

lying DWHA or DM may also need to be reconciled

to provide cleansed, integrated information. There-

fore, the four architectures (the Hub-and-Spoke, Data

Mart Bus, Centralised and Federated) have a similar

structure as the three-layered DWHA has. The in-

dependent DWHA extended from the concept of in-

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

608

Table 1: The data warehouse architecture distribution.

DWH Architecture Distribution

Hub and Spoke

Data Mart Bus

Centralised

Independent

Federated

Virtual

Distributed

Big DWHA

Data Lake

(Ariyachandra and Watson, 2008) 39% 26% 17% 12% 4%

(Matouk and Owoc, 2012) 35% 23% 20% 15% 7%

This paper 29% 9.3% 27% 8.7% 6.4% 2.9% 8.7% 3.5% 5.2%

dependent DMs has different situations: If it con-

sists of independent DMs mainly for individual de-

partment applications, then it may belong to tradi-

tional two-layered DWHA; If it is mixed with in-

dependent DWHs and DMs, it could belong to the

three-layered DWHA as it may also need to reconcile

data from different sources. Therefore, the relation-

ship between these two main classiﬁcations (Layer

Based Classiﬁcation and Component Based Classiﬁ-

cation) can be illustrated: Two-Layer ⊃ {Independent

[DM]}; Three-Layer ⊃ {Hub-and-Spoke, Data Mart

Bus, Centralised, Independent [DWH with DM] and

Federated}. By analysing the two main DWHA clas-

siﬁcations and collected DHWAs from literature, we

propose a new classiﬁcation model of DWHAs based

on their characteristics, which are structured in the top

of table 2 with bold font.

5.1 Discussions and Implications

Even DWHAs are classiﬁed in this model, there are

still some indistinct characteristics or similar compo-

nents which could confuse their differentiation. In or-

der to easily identify and explain the differences of

these DWHAs, three features and ﬁfteen sub-features

are chosen and discussed, which mainly derive from

generalised and modelled DWHAs in ﬁgure 1. Some

typical DHWAs collected from literature are selected

and discussed here as cases. The table 2 analyses and

summarises differences of these cases. The tick in

this table means that a certain DWHA has this sub-

feature or can address this sub-feature. For example,

the Virtual DWHA manipulates structured data and

then a tick is placed in the related cell. As can be seen

in this table, differences could be analysed from the

feature perspective. For example, most of DWHAs

cannot directly process unstructured data except Big

DWH and DL. Hence, if a company needs to anal-

yse a large amount of unstructured data directly, these

two architectures may be more suitable than others.

In order to distinguish the differences and better un-

derstand these DWHAs, some of them are analysed

together, as they have similar components or in the

same branches.

Virtual, Federated and Big Data: From ﬁgure 1,

the Virtual, Federated and Big DWHAs could contain

a logical DWH, but from table 2, the virtual DWHA

normally manages data with a single data schema in a

source system, while, the Federated and Big DWHAs

could extract data from multiple data sources. In ad-

dition, the Big DWHA store data using the distributed

ﬁle system (DFS); the Federated DWHA may extract

data from different sources and save data in traditional

database management systems.

Independent [DWH with DM], Centralised, Data

Mart Bus, Hub-and-Spoke and Federated: In the

Classiﬁcation Model, the Centralised, Data Mart Bus,

Hub and Spoke, and Federated DWHAs are cate-

gorised in the three-layer branch. The differences

mainly can be discerned in the ETL and DM fea-

tures. For example, the Data Mart Bus has separated

DMs but connected with conformed dimension tables;

the Hub-and-Spoke has DM(s) getting data from its

depended DWH, but they do not need to be strictly

linked by conformed dimension tables. The Indepen-

dent [DWH with DM] has DWH(s) and DM(s), but

they are independent.

Distributed, Big DWHA and Data Lake: The Dis-

tributed, Big DWHA and DL belong to the same

main branch. The data could be extracted, loaded

then transformed (ELT) in the Big DWHA , as the

Big DWHA has powerful computing platforms (e.g.

Hadoop) to manipulate data. In addition, its big DWH

(e.g. Hive) does not necessarily pre-load data, which

only needs to be informed where the data is stored in

DFS. However, data in the Distributed DHW should

be extracted, transformed then loaded into a DHW

with a predeﬁned schema (ETL). Besides, the DL

stores row data with detailed granularity, which ex-

tracts and loads (EL) row data from sources with less

pre-processing.

Analysis of Data Warehouse Architectures: Modeling and Classiﬁcation

609

Table 2: DWHA Classiﬁcation Model.

DWHA Classiﬁcation Model

Data Warehouse Architecture

Traditional Data

Oriented Architecture

Big Data Oriented

Architecture

One

Layer

Two

Layer

Three Layer

Big Synergetic

Virtual

Independent

[DM]

Independent

[DWH with DM]

Centralised

Data Mart Bus

Hub-and-Spoke

Federated

Distributed

Big DWHA

Data Lake

Data Type

Structured Data X X X X X X X X

Structured &

Unstructured Data

X X

Schema

Single X X

Multiple X X X X X X X X

ETL

Extract X X

Extract & Load X X X

Extract & Transform

Extract &

Load & Transform

Extract &

Transform & Load

X X X X

Data Mart

Conformed Data Mart

Independent Data Mart

X X

Dependent Data Mart

Unknown or Null

X X X X X X

Storage

Distributed X X X

Aggregated X X X X X X X

6 CONCLUSION

This paper conducts a quantitative study associated

with a systematical literature review on DWHAs to in-

vestigate them from the traditional data oriented and

big data oriented perspectives. 116 publications are

collected from literature and 73 disparate DHWAs are

modelled using Archimate. Based on these modelled

DWHAs and their characteristics, 9 representative

DWHAs are identiﬁed and summarised into a ”big

picture” which directly demonstrates their similarities

and differences. The number of these architectures

are counted and their distribution is generated by the

statistic technique. Furthermore, a new DWHA clas-

siﬁcation model is proposed to better identify, anal-

yse and compare DWHAs from featured and struc-

tural perspectives. For future work, the classiﬁcation

model would be reﬁned with more features to bet-

ter measure the performance or difference of these

DWHAs. Also, this model would be extended to ac-

commodate new technologies and DWHAs in further

research.

ACKNOWLEDGEMENTS

This publication is supported by the Science Founda-

tion Ireland grant SFI/12/RC/2289 to Insight Centre

for Data Analytics. INSIGHT is funded under the SFI

Research Centres Programme and is co-funded under

the European Regional Development Fund.

REFERENCES

Aliaga, M. and Gunderson, B. (2006). Interactive Statistics,

3rd Edition. Pearson.

Alsqour, M., K, M., and Owoc, M. L. (2012). A survey of

data warehouse architectures: preliminary results. In

InFedCSIS (pp. 1121-1126).

Ariyachandra, T. and Watson, H. (2010). Key organiza-

tional factors in data warehouse architecture selection.

In Decision support systems. 2010 31;49(2):200-12.

Ariyachandra, T. and Watson, H. J. (2006). Which data

warehouse architecture is most successful? In Busi-

ness intelligence journal, 11(1), p.4.

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

610

Ariyachandra, T. and Watson, H. J. (2008). Which data

warehouse architecture is the best? In Communica-

tions of the ACM, 51(10), p.146. ACM.

Ayman, A., Zaiane Osmar, R., and Ji, Y. (2001). Towards

framework for the virtual data warehouse. In British

National conference on databases (pp. 202-218).

Bla

c, G., Po

c, P., and Jak

c, D. (2017). Data ware-

house architecture classiﬁcation. In Information and

Communication Technology, Electronics and Micro-

electronics (MIPRO), 2017 40th International Con-

vention on (pp. 1491-1495). IEEE.

Bontempo, C. and Zagelow, G. (1998). The ibm data ware-

house architecture. In Communications of the ACM,

1;41(9):38-48. ACM.

Breslin, M. (2004). Data warehousing battle of the giants.

In Business Intelligence Journal.

Chaki, N. and Sarkar, B. (2010). Virtual data ware-

house modeling using petri nets for distributed deci-

sion making. In Journal of Convergence Information

Technology, 5(5), pp.8-21.

Devlin, B. (2014). Data lake muddies the waters

on big data management. In Available at:

https://searchbusinessanalytics.techtarget.com/feature/

Data-lake-muddies-the-waters-on-big-dat-

management [Accessed 15 Novmber 2018].

Devlin, B. and Cote, L. (1996). Data warehouse: from ar-

chitecture to implementation. Addison-Wesley Long-

man Publishing Co., Inc.,.

Dixon, J. (2010). Pentaho, hadoop and data lakes. In Avail-

able at: https://jamesdixon.wordpress.com/2010/

10/14/pentaho-hadoop-and-data-lakes/ [Accessed 15

Novmber 2018].

Ge, M. and Helfert, M. (2006). A framework to assess deci-

sion quality using information quality dimensions. In

11th International Conference on Information Qual-

ity, pages 455–466.

George, S. (2012). Inmon vs. kimball: Which approach is

suitable for your data warehouse. data warehous. In

pp.1-12.

Inmon, B. (2016). Data Lake Architecture: Designing the

Data Lake and Avoiding the Garbage Dump. Tech-

nics.

Inmon, W. (1997). What is a data warehouse? In Prism

Tech. Topic 1(1).

Katal, A., Wazid, M., and Goudar, R. (2013). Big data:

issues, challenges, tools and good practices. In Con-

temporary Computing (IC3), 2013 Sixth International

Conference on (pp. 404-409). IEEE.

Kimball, R. and Ross, M. (2013). The data warehouse

toolkit: the complete guide to dimensional modeling.

John Wiley & Sons.

Matouk, K. and Owoc, M. L. (2012). A survey of

data warehouse architectures—preliminary results. In

Computer Science and Information Systems (FedC-

SIS), 2012 Federated Conference on (pp. 1121- 1126).

IEEE.

Menon, S. (2005). Allocating fragments in distributed

databases. In IEEE transactions on parallel and dis-

tributed systems, 16(7), pp.577-585.

Modelling, A. (2018). Archi – open source

archimate modelling. In Available at:

https://www.archimatetool.com/ [Accessed 15

Novmber 2018].

Muijs, D. (2010). Doing quantitative research in education

with SPSS. SAGE Publications Ltd.

Pasquale, C., Ionescu, B., Ionescu, D., and Gurerero,

A. (2008). Virtual data warehouse architecture

for real-time webgis. In Virtual Environments,

Human-Computer Interfaces and Measurement Sys-

tems, 2008. IEEE Conference on. IEEE.

Pasupuleti, P. and Purra, B. (2015). Data Lake Development

with Big Data. Packt Publishing Ltd.

Patten, M. L. and Newhart, M. (2018). Understanding Re-

search Methods: An Overview of the Essentials. Tay-

lor & Francis, New York and London.

Pavlo, A., Curino, C., and Zdonik, S. (2012). Skew-aware

automatic database partitioning in shared-nothing,

parallel oltp systems. In Proceedings of the 2012 ACM

SIGMOD International Conference on Management

of Data (pp. 61-72). ACM.

Rangarajan, S., Liu, H., Wang, H., and Wang, C. (2015).

Scalable architecture for personalized healthcare ser-

vice recommendation using big data lake. In Service

Research and Innovation (pp. 65-79). Springer, Cham.

Research, C. (2014). Putting the data lake to work a guide

to best practices. In Teradata.

Robson, C. and McCartan, K. (2018). Real world research.

John Wiley & Sons.

Sahama, T. and Croll, P. (2007). A data warehouse architec-

ture for clinical data warehousing. In Proceedings of

the ﬁfth Australasian symposium on ACSW frontiers-

Volume 68. Australian Computer Society, Inc.

Scheibe, K. and Nilakanta, S. (2008). Dimensional issues

in agricultural data warehouse designs. In pp.263-278.

electronics in agriculture, 60(2).

Senapati, R. and Kumar, D. A. (2014). A survey on data

warehouse architecture. In International Journal of

Innovative Research in Computer and Communica-

tion Engineering.

Thusoo, A., Sarma, J., Jain, N., Shao, Z., Chakka, P., Zhang,

N., Antony, S., Liu, H., and Murthy, R. (2010a). Hive-

a petabyte scale data warehouse using hadoop. In

Data Engineering (ICDE), 2010 IEEE 26th Interna-

tional Conference on (pp. 996-1005). IEEE.

Thusoo, A., Shao, Z., Anthony, S., Borthakur, D., Jain, N.,

Sarma, J. S., Murthy, R., and Liu, H. (2010b). Data

warehousing and analytics infrastructure at facebook.

In Proceedings of the 2010 ACM International Con-

ference on Management of data (pp.1013-1020).

Yang, Q., Ge, M., and Helfert, M. (2017). Data quality

problems in tpc-di based data integration processes.

In International Conference on Enterprise Informa-

tion Systems (pp. 57-73). Springer, Cham.

Yang, Q. and Helfert, M. (2016a). Data quality for web log

data using a hadoop environment. In ICIQ 2016: 199-

208. MIT Information Quality ( MITIQ ) Program.

Yang, Q. and Helfert, M. (2016b). Revisiting arguments

for a three layered data warehousing architecture in

the context of the hadoop platform. In Proceedings of

the 6th International Conference on Cloud Computing

and Services Science, Volume 2. SCITEPRESS.

Analysis of Data Warehouse Architectures: Modeling and Classiﬁcation

611