Professional Documents
Culture Documents
Data Vault Modelling
Data Vault Modelling
Bachelor of Engineering
Information Technology
Thesis
31 March 2014
Abstract
The theme of this thesis is to prepare an introductory guide to Data Vault Modelling. The
Data Vault is a relatively new method of modelling enterprise data warehouses. The Data
Vault uses three core components to model an entire data warehouse. Thus it provides an
easy alternate solution for data warehousing.
The thesis project was conducted by researching books, scientific journal articles, profes-
sional blog posts and professional community discussions. By doing so scattered infor-
mation about Data Vault Modelling was gathered and analysed and compiled in to a
guidebook. This paper could be used as a quick guidebook for those interested in finding
information about Data Vault Modelling.
Data Vault is a relatively new solution to enterprise data warehousing. Although it intro-
duces a better way of modelling enterprise data warehouses, it still has limitations when it
comes to providing strict guidelines and handling unstructured text, semi-structured Doc
Style data, XML and other data types which are broadly known as Big Data. These data
types cannot be handled using the current approaches, tools and techniques; however
there is an ongoing research to make the Data Vault Model be able to handle such Big
Data. Thus this thesis includes only officially published articles about Data Vault.
1 Introduction 4
2 Theoretical Background 5
6 Conclusion 33
References 34
1 Introduction
Information plays a fundamental role in decision making in everyones life. In the enter-
prise world every piece of information from different sources of operation are described
as data. Data are the backbone of an enterprise and plays a major role in decision
making in different aspects regarding the enterprise. Choosing the proper way of the
data warehouse modelling method helps modelling a database layer where data are
stored and become available for retrieval for the purpose of decision making, data min-
ing and other activities.
There are several data warehouse modelling methods. The most common competing
options for modelling are either modelling with confirmed dimensions and an enterprise
data bus, or modelling with the database normalized. Although these methods have
their own advantages, they have issues when it comes to changes in the systems feed-
ing the data warehouse. Data cleansing for confirmed dimensions will lead to losing
information. A new way of data warehouse modelling, Data Vault came into case in
year 2000 and it was designed to minimize the impact of the issues mentioned above.
The purpose of this paper is to prepare an introductory guide to Data Vault modelling
basics. Since Data Vault modelling is a new concept of data warehouse modelling,
there are few published reference books. Most information regarding Data Vault model-
ling is scattered in the form of journals, newsletters, blogs and community discussions.
The goal of this thesis is to provide information about what Data Vault modelling is,
what core components it contains, how it can be modeled and used as an enterprise
data warehouse model.
5
2 Theoretical Background
Since the introduction of the Third Normal Form (3NF) in the early 1960s for on-line
transaction processing systems, the need for data warehouses was increasingly grow-
ing. Data warehousing, when used properly will enhance information managers of an
organization with timely information necessary to effectively make crucial business de-
cisions and solutions. [12, 23].Figure 1 shows the data warehousing timeline.
Figure 1.The Data Warehousing Timeline. Adapted from Kent Graziano (2011) [10, 8]
6
The term data warehouse is a generalized phrase that has taken different meanings
according to different experts in the field. According to Michael Brackett, a data ware-
house is a repository of consistent historical data that can be easily accessed and ma-
nipulated for decision support.[13, 268-269] W.H. Inmon who came up with the term
data warehouse initially defines it as a subject-oriented, integrated, time-variant, and
non-volatile collection of data in support of managements decision-making pro-
cess.[14, 2]
A data warehouse is a relational database designed for query and analysis. It usually
includes historical data derived from transaction data or other means of sources. In
addition to relational databases, a data warehouse environment may include an extrac-
tion, transportation, transformation, online analytical processing (OLAP) and data min-
ing capabilities, different client analysis tools and other applications that manage the
process of gathering and delivering data.[15,35-36].
There are different data warehouse characteristics set by William Inmon such as:
Subject-Oriented
Integrated
Non-Volatile
Time Variant. [15, 36]
Data warehouses are designed to help data analysis. The concept subject-oriented
depends on the change of data from application-oriented to decision support. Applica-
tion oriented data are detailed data based on functional requirements, whereas data for
data warehouses only include data for operating decision analysis. [14, 3-4; 15, 36]
One of the key elements for decision analysis to predict future matters about an organ-
ization is historical data. Historical data refers to changes over time and these changes
can be tracked by introducing a time element in the data warehouse. Time variance in
a data warehouse can be shown in several ways. Data warehouses contain data over
7
An RDBMS is a database system that organizes and accesses data as two dimension-
al rows and columns. In this case data are organized in such a way that related infor-
mation can be accessed using Structured Query Language (SQL). Although it is one
way of organizing data warehouse, RDBMS faces difficulty when it comes to supporting
complex analytical and decision making processes. The reason is that each time a
query is executed it must collect the data the query is seeking and that sometimes in-
volve going through millions of records. Due to this reason, RDBMS was not the best
choice for data warehousing projects that involved numerous complex queries. Thus it
has led new technologies to emerge to handle the case. [12, 26]
table with a single-part key that relates directly to the fact table. The fact table contains
the measures of interest and relates all the dimensions in a star-like structure by using
a multi-part key. Figure 2 shows an example of a fact table in MDDB.
Figure 2. Fact table in MDDB. Adapted from Robert W.Dill (1985) [12, 27]
In MDDB the fact table is used to traverse data across multiple attributes quickly. The
dimension tables contain the actual descriptive data. Data in MDDB are stored accord-
ing to common usage patterns of users in such a way that frequently accessed data
are summarized, preprocessed and made available for a user to query when needed.
Contrary to RDBMS that has to process the query dynamically each time a request is
made for data, MDDB is preferable for quick retrieval of predefined calculations and
results. [12, 27]
9
Regarding modelling and data logistics, data warehouse architectures can generally be
categorized as traditional, hybrid and modern. A traditional architecture includes infor-
mation factory and enterprise buses. On the other hand hub-and-spoke is an example
of a hybrid architecture. Modern architectures include Data Vault and Anchor Model-
ling. There are several other types of data warehouse architectures. Independent data
marts, data mart bus architecture with linked dimensional data marts, a centralized
data warehouse and federated data warehouse are among other architectures which
data warehousing literature suggests. Despite their variety, data warehouse architec-
tures contain common core components such as source systems (external or opera-
tional), data staging area, data presentation area and data access tools. Figure 3 ex-
plains the data warehouse architecture with common components.
10
Figure 3. Data warehouse architecture with common components. Adapted from Erike
Perjons [7, 1]
Source systems refer to different data sources that feed data into the data warehouse.
Data can be available in any format such as relational database, excel files or plain text
files. Data sources can generally be classified as operational, which is usually available
in the data warehouse operating systems and external, which includes data out of the
operating system. Operational data such as sales data, HR data, product data, market-
ing data, system data, and web server logs with a user browsing data combined with
external third-party data can all act as source systems. [8]
The data staging area is where data stands before it is transformed into a data ware-
house or data mart. It is often the complex part in the architecture and involves extract-
ing, transforming, loading and indexing activities. Extraction refers to reading and un-
derstanding the source data and copying the data needed for the data warehouse for
further manipulation and other activities such as transformation [7, 4-5;
8].Transformation involves the following activities:
11
The data presentation area usually refers to a series of integrated data marts. A data
mart is a flexible set of data based on granularity. In its simplistic form a data mart rep-
resents data from a single business process. The data presentation area is where in-
formation that reaches the users is available [7, 9; 8].Data accessing tools refer to ac-
tivities made by the end user to get data from the presentation area. These activities
include query/reporting, data mining and analysis.
12
Based on the data source that feeds the data mart, data marts could be categorized as
dependent, independent and hybrid. Dependent data marts get data from a central
data warehouse that already exists. As the name implies, independent data marts are
standalone systems which draw data directly from operational or external sources. Hy-
brid data marts can get data from operational systems or a data warehouse.
The key difference between dependent and independent data marts is the way data is
populated from the sources into the data mart. This process is called Extraction-
Transformation-Transportation (ETT). It involves moving of data from operational sys-
tems, filtering and loading it to the data mart. The ETT process in dependent data
marts is simplified due to the existence of formatted and cleaned data in the central
data warehouse. Therefore the ETT process in dependent data marts is a matter of
identifying the right set of data related to the specified subject area in the chosen data
mart. Independent data marts however should be dealt with all aspects of the ETT pro-
cess. The motivation behind the choice of different types of data marts depends on
improved performance and availability, better control and costs of time for solutions.
[16]
Like data warehouses, data marts can be categorized into two design architectures as
relational and multi-dimensional data marts. The choice between these architectures
depends on the type of analysis to be performed and the type of data to be analyzed.
[12, 32]
Multi-dimensional data marts (MDDM) which maintain a large amount of numeric data
(such as sales data) are used to suggest different analytical ways on the same data.
Once the data are loaded from data warehouse or external sources, it should be avail-
able in a structured framework to facilitate analysis. MDDM are efficient for analyzing
13
numeric data. The method of analytical processing for decision support is known as
On-line Analytical Processing (OLAP). [12, 32]
The relational data marts allow more general-purpose decision-making tools than
MDDM. The relational data mart uses a type of analysis process called Relational On-
line Analytical Processing (ROLAP).It supports a range of purposes for numeric and
textual data. [12, 32]
The main difference between a data warehouse and a data mart is in the size of data
maintained. The range of subject areas to deal with and the number of sources the
data are drawn from create size differences that a data warehouse and data mart could
handle. Although the difference in size is visible in most cases, there exist more differ-
ences in the application and implementation. Table 1 shows the main differences be-
tween data warehouses and data marts.
Table 1. Differences between data warehouses and data marts. Adapted from Robert
W.Dill (1985) [12, 30]
Every day organizations generate and record massive volumes of data. There exist
many business intelligence tools which could extract these data for the purpose of bet-
ter business analysis and decision making. However business environments and con-
sumer behaviors are altering rapidly and thus the market is so unpredictable that re-
quires the use of all forms of data that are available. [22, 2]
With the huge growth in Internet-based businesses and availability of many social net-
works, many enterprises now record terabytes and petabytes of data in their data
warehouse. Even though there exist enough business intelligence tools for analytical
reporting purposes, these abundantly available data are not considered on analytical
purpose. Enterprises now need better business intelligence tools to utilize the full po-
tential of Big Data that are available. This in turn will help them take analytical deci-
sions with innovative ideas and firm business decisions in real time. [22, 2]
On a daily basis businesses today create 2.5 quintillion bytes of data which is so much
that 90% of the data in the world today has been created in the last few years alone.
These data generates from everywhere: sensors used to collect different information,
posts and conversations in social media sites, digital pictures and videos, transaction
records and cellphone signals among others. These data are Big Data. Thus Big Data
are beyond the matter of size. It is an opportunity to find new insights about emerging
types of data and content to make businesses more agile and to solve potential prob-
lems that are considered beyond the reach of business intelligence. [22, 3]
15
Many companies have traditional data warehouses that were not designed to handle
big data. However nowadays there are ways to evolve these traditional data ware-
houses into 'analytics warehouses' so that processing structured and unstructured data
could be possible. Traditionally data warehouses were designed to handle structured
data, not the type of unstructured data from social media, mobile devices, web traffic
and other sources that are feeding data into enterprises now. However vendors are
building new generation data warehouses with statistical capabilities for performing
analytics and forecasting. New generation data warehouses are designed to support
structured and unstructured data. Thus enterprises will be capable of having all-
rounded views of their operations in order to make better decisions about the future.
[23]
A distributed file system, which sits between source data systems and the data
warehouse so that it collects, aggregates and processes massive volumes of
unstructured data and stages it into the data warehouse.
Structured and unstructured data from backend systems which can be brought
into the data warehouse in real and near-real time.
Engines that use predictive modeling techniques to identify patterns in big data
and support real-time decision making. [23.]
Many companies are now considering their traditional data warehouse infrastructures
to integrate big data technologies, analytics technologies and other emerging business
intelligence technologies to exploit the full potential of data in their data warehouse for
better analysis and decision support.
17
The research and development on the Data Vault modelling approach began in the
early 1990s by Dan Linstedt and was completed around 1999.The following three years
the Data Vault design was tested, refined and deployed to selected adopter sites. In
addition to that a series of articles on Data Vault modelling were released by Dan Lin-
stedt. In 2002 the Data Vault architecture was reviewed to meet the needs of the data
warehouse in industries. In 2003 teaching the Data Vault modelling techniques began
to the public. According to its creator Dan Linstedt, the Data Vault is a detail-oriented,
historical tracking and uniquely linked set of normalized tables that support one or more
functional areas of business. It is a hybrid approach composed of the best breed be-
tween 3NF and star schema. The Data Vault is designed to be flexible, scalable, con-
sistent and adaptable to the needs of the enterprise.[9, 9]
The Data Vault focuses on data integration, traceability, auditability and resilience to
change. It is particularly strong in tracing of data warehouse data back to their source
system ensuring a complete audit trail. The Data Vault Model can easily be adapted to
varying requirements and it is highly scalable. It is based on principles that are related
to normalized data models in such a way that the Data Vault model consists of struc-
tures which resemble the traditional definitions of star schema and 3NF including di-
mensions, many- to- many relations and standard table structures. However there is a
difference in relationship representations, field structuring and granular time-based
data storage. With the volume of information warehouses have to store exponentially
increasing every year, 3NF and star schema have issues when it comes to scalability,
flexibility and granularity of data. The Data Vault is modeled to solve these bottlenecks
of data warehouses with its core components, the hubs, links and satellites. [19, 8]
The Data Vault can be applicable in Dynamic Data Warehousing where dynamic auto-
mated changes are made to both process and structure within the warehouse. It allows
users to explore the structures of the data warehouse without losing the content. In
addition, the Data Vault could be applied in In-Database Data Mining by allowing data
mining tools to use the historical data and to fit with the functions of artificial intelli-
gence. [19, 9]
18
The foundation the Data Vault architecture is rooted in reduction of redundancy. The
Data Vault architecture fits with the traditional Bill Inmons Corporate Information Facto-
ry (CIF) approach in such a way that the Data Vault plays the role of centralized enter-
prise data warehouse (EDW).In this case the Data Vault provides data to star schema
data marts, report tables and exploration marts.[2,2] Figure 4 explains the Data Vault
EDW architecture.
Figure 4. Data Vault EDW Architecture. Reprinted from Enterprise Wide Business Key
[18, 17]
Figure 4 represents high level Data Vault architecture in enterprise data warehouse.
Business rules in BDW (Business Data Warehouse) layer could be applied in the same
or separate area of the EDW (Enterprise Data Warehouse) layer. Data marts can draw
data from the BDW or EDW layers as appropriate. Data in the Stage and Data Mart
area are not kept but rather overwritten, so the Stage and Data Mart areas are not per-
sisted. Components in the EDW and BDW layers are commonly persisted. The EDW,
which is the core historical data repository, consists of Data Vault modeled tables. The
tables in the EDW are composed of hubs, links and satellites which are considered to
be the fundamental elements of the Data Vault model. The EDW layer is then com-
prised of the Data Vault Model where raw granular data sets are stored. The OLAP
(On-line Analytical Processing) cubes draw data from data marts which could be used
for decision-making, data mining or artificial intelligence tools. [3, 14]
19
There are three core entity components in the Data Vault model. The hub, link and sat-
ellite entities. The Data Vault design is focused on the functional areas of businesses
with hubs representing the primary business keys. The link entities associate two or
more business keys and they are commonly referred as transactions. The satellite enti-
ties provide a descriptive context of the primary keys or their associations. [1; 2, 3]
3.2.1 Hubs
Hubs are defined by a unique list of business keys which represent core business con-
cepts such as customer, vendor, sale or product. These business keys are vital to
businesses to track, locate, and identify their information. The hub table being a busi-
ness key recording device is established the first time a new instance of the business
key is introduced to the EDW. They also provide additional technical information such
as load time stamp, record source and sequence number also knows as the surrogate
key. [2, 5]
The purpose of a hub is to trace the first time the Data Vault experiences the business
key arrival in the warehousing load and where it came from. It provides a soft-
integration point of raw data that have the same semantic meaning to the source sys-
tem. The hub key allows a business to track their information across their business line,
which in turn provides a consistent view of the current state of their systems. [2, 6; 3, 4]
The hub entity table structure consists of a minimum of the required elements such as
a surrogate sequence id, a business key, a load date / time stamp and a record source.
Hubs are parents to all other tables. Figure 5 shows the basic hub entity table structure
for an example business concept customer.
20
The Hub is logically an independent key and none of its part is dependent on another
business key. In some cases there might exist two business keys. In these cases one
of the keys is the surrogate key of the other business key. The surrogate key cannot
become a Hub when there is another independent key. Thus we can choose to ignore
surrogate keys altogether. The surrogate key can be considered as a technical key that
can serve as a Hub when there is no other candidate independent key. [2, 6]
21
3.2.2 Links
Link entities are the glue that pulls together any relationships between two or more
business keys and sometimes other Links. They are established the first time a new
unique association is presented to the EDW. Links act as the flexible component of the
Data Vault model and are created when business keys interact. The Links contain the
surrogate sequence ids from the Hubs and Links that it associates to, warehouse se-
quence id, a load date / time stamp and a record source. Like the Hubs, Links contain
no descriptive information. [2, 8; 3, 4]
The job of the Link is to capture and record the past, present, and future associations
of data elements at the lowest possible grain. It provides flexibility and scalability to the
Data Vault model by allowing changes, modification and adaption to the structure over
time without losing auditability and compliance. [2, 8]
Data Vault modelling constructs form a link any time when there is a one-to-one, one-
to-many, many-to-one, or many-to-many relationship between business keys or data
elements. Although these relationships come with their own benefits and drawbacks,
allowing many-to-many relationships within business keys provide the following bene-
fits: [2, 9; 5, 7]
A many-to-many relationship enables rapid data changes on a massive scale with little
to no impact on both existing data sets (referred as history) and existing processes
(commonly known as load and query).It makes businesses change at the speed of a
business and become more agile and responsive to handle those changes. In general
by modelling the Links as a many-to-many relationship we can accomplish the above
goals. [2, 9]
22
The Link entity table structure consists of the basic required elements such as surro-
gate sequence id, multiple business sequence keys from the Hubs and other related
Links, load date stamp, and record source. Depending on business requirements for
query purposes, items such as last seen date, encryption key and other metadata ele-
ments may be added. Figure 6 shows the Link table structure based on an example
business concept Customer and Hub table in Figure 5.
Figure 6. The Link table structure. Reprinted from Dan E.Linstedt [3, 4]
The above structure contains Hub surrogate keys from H_Customer in Figure 5 and
H_Cust_Class which is another business concept to which relationship would be ex-
tended. Generally the links contain the following attributes: [2, 9]
Hub surrogate keys: business sequence keys which establish the linkage back
to the Hub unique entity.it is considered a mandatory key for the links.
Link surrogate key: an optional key but it could be considered as a mandatory
primary unique key in case of Link associated to other Links. It is highly im-
portant in allowing Satellite data to relate to Links in flexible and dynamic man-
ner.
Loading timestamp: a mandatory key for the links and it audits data recording
when an association was first introduced.
Record source: a mandatory key used for traceability and integrations purpos-
es. [1]
23
3.2.3 Satellites
The Satellite entity table structure consists of basic required attributes such as surro-
gate sequence id from the Hubs or Links table, load date stamp and record source.
The primary key of the Satellite is a two part key combining the Sequence key of the
parent (Hub or Link) with the Load Date time stamp. Real-time data can easily be add-
ed in to Satellites by adding millisecond timer and without creating duplicate primary
keys. Figure 7 shows the Satellite entity table structure based on a business concept
Customer which is also used for Hubs and Links structure tables above.
24
Figure 7. The Satellite entity table structure. Reprinted from Dan E.Linstedt [3, 5]
Components of the Satellite table such as Data/Time Stamp and Record source are
similar to the Hub and Link tables entities. These entities are very important in tracking
history of data by exposing the date and time the data was entered to the system and
where the data came from. Some of the entities figure 7 contains are explained below.
Hubs, Links and Satellites are the building blocks for the Data Vault in an EDW. The
Hubs and Links together create the backbone or skeletal structure of the model and the
Satellites provide all the context and all the history (changes over time).The skeletal
structure represents a one-to-one relationship with the core business concepts (Hubs)
and their business associations (Links).Adding Satellites to the backbone structure will
form a complete data vault model and all the core constructs together represent all
integrated data from the organization.
The process of modelling with the Data Vault involves several steps. The first step
though is to identify business concepts for the given subject area. Identifying business
concepts help to establish EWBK (enterprise wide business key) for Hubs, which in
turn will lead to defining or modeling the Hubs. Once the Hubs are defined, the next
step is identifying the natural business relationships between the Hubs. Identifying rela-
tionships between Hubs help to model Links. Now the skeletal structure could be
formed using the modeled Hubs and Links. The next step is to gather context attributes
to define keys and establish criteria to model satellites, which will provide descriptive
information to the Hub and Link constructs. [3, 6]
There are several rules which should be considered when modeling a Data Vault.
Some of the reference rules for Data Vaults are the following and the detailed rules are
provided in appendix 1:
A Hub table always migrates its primary key outwards: Hub keys cannot migrate
into other Hubs, which means there should not be parent or childlike Hubs.
Hub-to-Hub relationships are allowed only through a link structure: Hubs must
be connected through links and more than two Hubs can be associated through
links.
Recursive relationships are resolved through a link table.
A Link structure must have at least two foreign key relationships: Links must
have at least two Hubs connected to them in order to be instantiated.
A Link structure can have a surrogate key representation: Surrogate keys may
be utilized for Hubs and Links but not for Satellites.
A Link structure has no limit to the number of Hubs it associates.
A Link-to-Link relationship is allowed.
26
The backbone structure (Hub and Link) and Satellite combine to form a Data Vault. A
Data Vault can be as small as a single Hub with one Satellite, or it could be as large as
the business scope permits. The Data Vault allows scope modification. Neither scala-
bility nor granularity of the information is not an issue for scope modification. A Data
Vault model with an example Customer business concept is provided in Figure 8.
Figure 8. Complete Data Vault Model. Reprinted from Dan E.Linstedt [3, 5]
Figure 8 shows how simple a Data Vault model could be constructed, the connection
between the skeletal structure and their descriptive information. A full example of a
sample Data Vault model is provided in appendix 2.
27
Loading a data warehouse is the key activity of data processing. It is performed in dif-
ferent processing modes. A processing mode is the way in which data are grouped and
handled to the warehouse for the purpose of generating reports, computation, analysis
and other outputs. Traditionally data warehouses are loaded with data from operational
systems in a batch mode. The Batch mode (batch processing) is a method in which a
number of similar transactions to be processed are batched (grouped together) as a
single unit at predetermined times. These accumulated transactions are then presented
to the data warehouse at regular intervals of time. It means that data in the warehouse
become available as extracted file on nightly, weekly or monthly schedule after the sys-
tem has transformed and cleansed it. As a result data warehouses do not contain to-
days data for decision support and this caused problems in a fast changing business
decisions. [19, 3]
There are several objectives to be achieved when a business chooses among different
data warehouse loading mechanisms. The following are some of the goals and objec-
tives for loading the Data Vault. [20]
Hubs record the first time the business key reached the Data Vault. They do not record
additional loads or other changes in data. Before loading a hub to the warehouse it is
always advisable to check if it already exists in order to avoid information duplication
and broken load process. Loading a hub could be simple in consideration that all the
hub business keys must be unique and one-to-one with surrogate keys, hub keys are
not time-based, hub keys do not support duplicates and hub keys are defined at the
same semantic layers. Identifying the master system where most business keys are
generated is the first step of loading a hub. These keys in the staging table that are
targeted to the hub are loaded as distinct or unique list and provided a surrogate key at
load time. [21]
The process of loading is repeated for each source system with a hub key field. The
field used to populate the hub key may not have any relation with foreign key in the
source system design. If a key exists, it must then be placed in the Hub. Information
with no key can be attached to a hub row called zero row key or UNKNOWN, so that
the data is not lost during the load of the Satellites. [21]
Hub duplication can cause issues ranging from query problems to complicated model-
ing problems. Thus it is recommended to avoid loading Hub keys with duplicates. How-
ever in the case of bad data, including the record source as part of the business key
could offer uniqueness to the Hub keys. In actual batch loading processes, all the hubs
can and should be loaded in parallel so as to produce a highly scalable architecture
with more data and more Hubs. [21]
In a real-time loading system the business keys almost always arrive with transactions.
In real-time business keys should be provided by the system to help the source system
attach the data to its own business processes. Real-time hub loading follows the same
cycle as the batch hub loading in such a way that it inserts the key if it does not already
exist. In addition to this architectures that handle millisecond feeds use a millisecond
time-stamp hub as their key. [20]
30
Link tables without surrogate keys follow the same loading process as Hubs. However,
since most enterprise data warehouses today use surrogate key structures, the load
process in the Link tables is extended to locate Hub rows containing surrogate keys.
Loading the Link tables consist of, formulating a unique list of all the business keys that
build up the relationship for the Link, locating each of the Hubs relating surrogate keys
and inserting the key with a new Link surrogate if it does not already exist in the target.
If the key already exists, it will be filtered out from the feed. [21]
The objective of loading the Links is to avoid duplicates across the business key struc-
tures in such a way that each relationship or intersection must be unique. Duplicates of
business keys in Link tables usually cause a poor definition of grain (meaning of single
row in the table) or mis-modeled Link entity. If there is a grain problem in the Link table
model in the Data Vault, the process cannot find an old satellite record to end date
even though it exists. To avoid this Links must be designed to fit the grain. [21]
Link table loads are considered successful when all relationships between business
keys have been made. In case of missing or null business keys in links, load zero
keys and tie the relationships to unknown Hub records. Thus data loading will be con-
tinued into the model without breakages. All the relationships have to be captured to
provide a full audit trial capability and trace the information back to its source system.
Like Hub loading, all Link loads can be run in parallel. [21]
Link loading in real-time could be easy as the Hubs match keys to business keys and
this matches are built into a new staging surrogate table. Thus Link load time is re-
duced. Millisecond feed transactions are kept with millisecond hub stamp and are
joined across the links via a process that run once a second or few seconds to build the
associations. Since the information in real-time arrives so quickly, introducing parallel-
ism to the process can avoid complexity out of the direct loading path. [20]
31
The main objectives of Satellite loading are to load delta changes and to split the work
by type of data or rate of change. The process of batch loading on a Satellite involves
gathering the link or hub surrogate key, selecting rows that have changed, joining to a
current load-date table and comparing the most recent data in the staging area to the
Satellite. Sometimes Satellites require Link PKs (Link sequence number) so loading
process also involves putting the new load-date into the PK of the Satellite. Finally
track the rows and insert the new Satellite attributes as simple as possible to end-date
old rows in the next pass and to update bridge tables. [20]
Establishing a unique list of Satellite attributes avoids loading duplicate attribute rows
into the Satellite target.it is suggested to have default values as part of the process to
utilize default data sets for some columns. There must be either the Hub key or the
Link parent key to attach the Satellite row properly. If parent key does not exist, loading
process must be stopped and design should be fixed for reuse. Furthermore compari-
son of the columns of the data coming in with the most recent Satellite data should be
made to ensure not to load duplicate data. Using end-dating process can make the
loading process fully restartable and scalable. The most current Satellite record that
has not been end-dated will be end-dated during the run. End-dating Satellite helps to
identify which data is active. [21]
32
As a hybrid data modelling approach Data Vault arguably consists of the best model-
ling solutions from the Third Normal Form (3NF) and Dimensional Modelling. However
when it comes to applicability Data Vault has its downsides. Table 2 explains a select-
ed list of merits and demerits of the Data Vault as an Enterprise Data Warehouse.
Table 2 Merits and Demerits of the Data Vault as an EDW. Data gathered from Data
Vault Overview [17, 16]
Merits Demerits
completely auditable architecture requires highly structured architect role
highly adaptable to (business) changes performance issue due to scalability
designed and optimized for EDW additional storage of data
scalability(provides multi-terabyte stor- lacks strict or formal implementation
age) guidelines
supports real-time processing requires business analysis to be firm
can be incrementally built and easily ex- not conductive to OLAP processing
tended
supports data mining and artificial intelli- introduces many joins and tables in que-
gence ries
supports flexible / extensible development not intended for end user access(such as
BI tools and OLAP)
The Data Vault as a hybrid data warehouse solution allows scaling to a large extent
which leads to more joins and tables in queries. In other words tables need to be joined
to achieve the result when making analysis or building a report. Although the concept
of the Data Vault is well documented there exists no strict implementation guideline,
which leads different people to apply different views on implementing it. Generally the
data warehouse layer contains a copy of data which could be considered as redundant.
Due to this we need additional storage area to handle data.
33
6 Conclusion
Data Vault modeling is very important when modeling a data warehouse. It contains a
wide range of modelling features for enterprise data warehouse projects. One of the
key features of Data Vault modeling is its ability to adapt to changes in requirements
both in sources and data marts. Using this modeling technique an enterprise could be
able to get the benefits of building dynamic data warehousing depending on dynamic
automated changes to the process and structure within the warehouse. Moreover it
helps to explore warehousing on which users could play with the data warehouse struc-
tures without losing any content.
Another benefit of using Data Vault modelling in an enterprise is data mining. The Data
Vault model allows data mining tools to make use of the historical data and fit the struc-
ture with the function of artificial intelligence. In addition to that it has an ability to rapid-
ly link and adapt to the structures of an external information source and create useful
data in the data warehouse without losing an existing content.
In contrast to a wide range of benefits, using Data Vault modeling has its own business
and technical issues. Data in the Data Vault are non-user accessible. Data Vault mod-
eling requires more up-front work even though it is for long-term benefits. Data in Data
Vault are not cleansed or its quality is not checked. Data Vault model introduces many
database join operations and it is not conductive to an OLAP (online analytical pro-
cessing) processing. Thus it requires a firm business analysis.
With a good range of benefits and limited disadvantages the usage of the Data Vault
approach is growing globally. Currently there are more than 500 companies which exist
in different parts of the world and which are using the Data Vault. IBM, Intel, Microsoft
Corp., Teradata Corporation are among the technology giants which use the Data
Vault. In general with the Data Vault team actively working on bringing the best model-
ing approaches, rules and practices to meet the industrial needs for data warehousing,
using the Data Vault could be seen as an up-to-date alternative approach for data
warehousing.
34
References
1. Dan E. Linstedt. Data Vault Series 1 - Data Vault Overview [Online]. [Accessed
2013 Dec 15];
URL: http://www.tdan.com/view-articles/5054/
2. Kent Graziano, Dan Linstedt. Introduction to Data Vault Modeling [e-Book]. [Ac-
cessed 2014 Jan 3];
URL:http://kentgraziano.files.wordpress.com/2012/02/introduction-to-data-vault-
modeling.pdf
3. Dan E. Linstedt. Data Vault Modeling Guide [e-Book]. [Accessed 2014 Mar 31].
URL:http://hanshultgren.files.wordpress.com/2012/09/data-vault-modeling-
guide.pdf
4. Jonathan Shirey. Data Vault: Hubs, Links, and Satellites with Associated Loading
Patterns [Online]. [Accessed 2014 Jan 19].
URL:http://makingdatameaningful.com/2012/09/18/data_vault-
hubs_links_and_satellites_with_associated_loading_patterns/
5. Martino Adriano. Further in Agility with Data Vault [e-Book]. 2013 05 [Accessed
2014 Feb 10];
URL:http://www.trivadis.com/uploads/tx_cabagdownloadarea/Further_in_agility_wit
h_data_vault_01.pdf
6. Ariyachandra HJW, Thilini. Data Warehouse Architectures: Factors in the S e lec-
tion Decision and the Success of the Architectures. Terry College of Business Uni-
versity of Georgia [e-Book]. 2005 Jul [Accessed 2014 Feb 19];
URL: http://www.terry.uga.edu/~hwatson/DW_Architecture_Report.pdf
7. Erik Perjons. DW Architecture and Lifecycle [e-Book]. [Accessed 2014 Jan 19].
URL: http://people.dsv.su.se/~petia/is5/Lectures/F4.pdf
8. Data Warehouse Architecture [Online]. [Accessed 2014 Jan 25];
URL:http://www.1keydata.com/datawarehousing/data-warehouse-architecture.html
9. Kent Graziano. (OTW13) Agile Data Warehousing: Introduction to Data Vault Mod-
eling [Online]. 2013 [Accessed 2014 Jan 30].
URL: http://www.slideshare.net/kgraziano/agile-data
10. Kent Graziano. Why Data Vault? [Online]. 2011 [Accessed 2014 Jan 25].
URL: http://www.slideshare.net/kgraziano/why-data-vault
35
11. Oracle Corporation. Oracle9i Data Warehousing Guide [Online]. [Accessed 2014
Jan 30]; Release 2 (9.2).
URL: http://docs.oracle.com/cd/B10501_01/server.920/a96520/concept.htm
12. Robert W. Dill LC Unites States Navy. Data Warehousing and Data Quality for a
Spatial Decision Support System [e-thesis]. 1985 [Accessed 2014 Feb 20].
URL: https://archive.org/stream/datawarehousingd00dill#page/n7/mode/2up
13. Brackett, Michael H. The Data Warehouse Challenge. New York, NY: John Wiley &
Sons, Inc.; 1996.
14. Inmon, William H. and Hackathorn, Richard D. Using the Data Warehouse. New
York, NY: John Wiley & Sons, Inc.; 1994.
15. Paul Lane. Oracle Database Data Warehousing Guide [e-Book]. Release 2 (10.2).
500 Oracle Parkway, Redwood City, CA 94065: Oracle Corporation; 2005 [Ac-
cessed 2014 Mar 5].
URL: http://docs.oracle.com/cd/B19306_01/server.102/b14223.pdf
16. Oracle Corporation. Oracle8i Data Warehousing Guide[Online]. [Accessed 2014
Mar 10]; Release 2 (8.1.6).
URL: http://docs.oracle.com/cd/A87860_01/doc/server.817/a76994/marts.htm
17. Empowered Holdings, LLC. Data Vault Overview [Online]. 2011 [Accessed 2014
Mar 12].
URL: http://www.slideshare.net/dlinstedt/data-vault-overview
18. Genesee Academy, LLC. Enterprise Wide Business Key: Aligning the Business w/
EDW & MDM. 2010 [e-Book] [Accessed 2014 Mar 20];
URL:http://www.enterpriseiq.com.au/documents/pres/DW_BI_2.0_Summit_D1_S6_
Hans_Hultgren_2010_Genesee.pdf
19. Oracle Corporation. Best Practices for Real-time Data Warehousing [e-Book]. 2012
Aug [Accessed 2014 Mar 21];
URL:http://www.oracle.com/technetwork/middleware/data-integrator/overview/best-
practices-for-realtime-data-wa-132882.pdf
20. Dan Linstedt. Data Vault Loading Specification v1.2 [Online]. 2010 May 13 [Ac-
cessed 2014 Mar 24];
URL: http://danlinstedt.com/datavaultcat/standards/data-vault-loading-specification-
v1-2/
21. Dan E. Linstedt. Data Vault Series 5 - Loading Practices [Online]. 2005 Jan 1 [Ac-
cessed 2014 Mar 25];
URL: http://www.tdan.com/view-articles/5285
36
22. Lt. Dr. Santhosh Baboo & P. Renjith Kumar. Next Generation Data Warehouse De-
sign with Big Data for Big Analytics and Better Insights. Global Journal of Computer
Science and Technology Software & Data Engineering [online].2013 [accessed
2014 April 9]
URL: https://globaljournals.org/GJCST_Volume13/3-Next-Generation-Data.pdf
23. The Future of Data Warehouses in the Age of Big Data, Deloitte CIO Journal Editor
[online]. July 17, 2013.[accessed 2014 April 9]
URL: http://deloitte.wsj.com/cio/2013/07/17/the-future-of-data-warehouses-in-the-
age-of-big-data/
Appendix 1
37