Whitepaper Cisco Datavault

Data Vault and Data Virtualization:

Double Agility
A Technical Whitepaper

Rick F. van der Lans
Independent Business Intelligence Analyst
R20/Consultancy
March 2015
Sponsored by

Copyright © 2015 R20/Consultancy. All rights reserved. Cisco and the Cisco logo are trademarks or
registered trademarks of Cisco and/or its affiliates in the U.S. or there countries. To view a list of Cisco
trademarks, go to this URL: www.cisco.com/go/trademarks. Trademarks of companies referenced in this
document are the sole property of their respective owners.

Table of Contents

1 Introduction 1
2 Data Vault in a Nutshell 2
3 The SuperNova Data Modeling Technique 8
4 Rules For the Data Vault Model 10
5 The Sample Data Vault 12
6 Developing the SuperNova Model 14
7 Optimizing the SuperNova Model 25
8 Developing the Extended SuperNova Model 28
9 Developing the Data Delivery Model 30
10 Getting Started 32
About the Author Rick F. van der Lans 33
About Cisco Systems, Inc. 33
About Centennium 33

Copyright © 2014 R20/Consultancy, all rights reserved.

Data Vault and Data Virtualization: Double Agility 1
1 Introduction

The Problems of Data Vault – Data Vault is a modern approach for designing enterprise data warehouses.
The two key benefits of Data Vault are data model extensibility and reproducibility of reporting results.
Unfortunately, from a query and reporting point of view a Data Vault model is complex. Developing
reports straight on a Data Vault‐based data warehouse leads to very complex SQL statements that almost
always lead to bad reporting performance. The reason is that in such a data warehouse the data is
distributed over a large number of tables.

The Typical Approach to Solve the Data Vault Problems – To solve the performance problems with Data Vault,
many organizations have developed countless derived data stores that are designed specifically to offer a
much better performance. In addition, a multitude of ETL programs is developed to refresh these derived
data stores periodically. Although such a solution solves the performance problems, it introduces several
new ones. Valuable time must be spent on designing, optimizing, loading, and managing all these derived
data stores. The existence of all these extra data stores diminishes the intended flexibility that
organizations try to get by implementing Data Vault. Philosopher and statesman Francis Bacon would
have said: “The remedy is worse than the disease.”

The New Approach to Solve the Data Vault Problems – An alternative approach exists that avoids the
performance problems of Data Vault and preserves its key benefits: reproducibility and data model
extensibility. This new approach is based on data virtualization technology1 together with a data modeling
technique called SuperNova. With this approach a data virtualization server replaces all the physical and
derived data stores. So, no valuable time has to be spent on designing, optimizing, loading, and managing
these derived data stores, and therefore reproducibility of reporting results and data model extensibility is
retained. Also, because new derived data stores are completely virtual, they are very easy to create and
maintain.

The SuperNova Modeling Technique – Applying the data modeling technique called SuperNova results in a set
of views defined with a data virtualization server plus (very likely) one derived data store designed to
support a wide range of reporting and analytical needs. SuperNova is a highly‐structured modeling
technique. It has been designed to be used in conjunction with Data Vault.

The Whitepaper – This whitepaper describes the SuperNova modeling technique. It explains step by step
how to develop an environment that extracts data from a Data Vault and how to make that data available
for reporting and analytics using the Cisco Information Server (CIS). Guidelines, do’s and don’ts, and best
practices are included.

Target Audience – This whitepaper is intended for those IT specialists who have experienced the problems
with Data Vault as discussed in this introduction or who want to avoid these problems. In addition,
general knowledge of data warehouses, star schema design, and SQL is expected.

1
R.F. van der Lans, Data Virtualization for Business Intelligence Systems, Morgan Kaufmann Publishers, 2012.

2 Data Vault in a Nutshell

This section contains a short introduction to Data Vault and the accompanying terminology. The term
Data Vault refers to two things. On one hand, it identifies an approach for designing data bases and, on
the other hand, it refers to a database that has been designed according to that approach. Both are
described in this section. For more extensive descriptions we refer to material by Dan Linstedt2 and Hans
Hultgren3.

The Data Vault – A Data Vault is a set of tables designed according to specific rules and meets reporting
and analytical needs of an organization. The tables together generally form an enterprise data warehouse.

Each table of a Data Vault is normalized4 and according to the third normal form, but if the entire set of
tables is considered, the tables are, what is called, over‐normalized or hyper‐normalized. Data that would
normally be stored together in one table when classic normalization is applied, is distributed over many
separate tables in a Data Vault. For example, an object and all its attributes are not stored together in one
table, but distributed over several tables. Also, a one‐to‐many relationship between two objects is not
implemented as a foreign key in one of the two tables, but as a separate table.

Another characteristic of Data Vault tables is that they all have particular metadata columns that indicate,
for example, when data is loaded, from which source system the data comes, when the data was last
changed, and so on. Metadata columns do not represent user information needs, but are needed to
process data correctly.

The Data Vault Approach – The Other Side of the Coin – As indicated, the term Data Vault also refers to a
specific modeling and design approach called the Data Vault Approach (in this whitepaper). It describes
how information needs are to be modeled as a set of extensible tables that, together, form a Data Vault.
In addition, it describes and prescribes how to unload data from data source systems, how to transform it,
how to cleanse incorrect data, and how to load it into a Data Vault.

The two, the database and the approach, go hand in hand. They are two sides of the same coin. It’s
strongly recommended to apply the data vault approach when a Data Vault has to be designed. Also,
when the approach is applied, the resulting database should represent a Data Vault.

The Tables of a Data Vault in Detail – The tables of a Data Vault are classified as hubs, links, and satellites.
Hubs are tables that represent business objects, such as product, customer, airport, aircraft, and so on.
Link tables represent relationships between hubs. Examples of link tables are flight (linking the hubs
aircraft and airport) and purchase of a product (linking the hubs product and customer). Satellites are
tables containing the attributes of hubs or links. Each satellite may contain one to many attributes. For
example, a satellite of the hub airport may contain the attribute called number of runways, the employee
hub may contain the monthly salary, and the customer hub may include the delivery address. An example
of a link satellite is the duration of a flight belonging to the link table called flight. Hubs and links can have
zero or more satellites. So, all the data stored in a Data Vault is stored in one of these three types of
tables: hubs, links or satellites. Figure 1 contains an example of a data model developed for a Data Vault‐
based data warehouse.

2
Dan E. Linstedt, Data Vault Series 1 ‐ Data Vault Overview, July 2002, see http://www.tdan.com/view‐articles/5054/
3
H. Hultgren, Modeling the Agile Data Warehouse with Data Vault, Brighton Hamilton, 2012.
4 th
Chris J. Date, An Introduction to Database Systems, 8 edition, Addison‐Wesley, 2003.


Figure 1 An example of a data vault data model.


In Figure 1, the tables whose names start with an H are hubs, with an L are links, and with an S are
satellites. The tables H_AIRCRAFT, H_AIRPORT, H_CARRIER, and H_FLIGHT_NUMBER form the hubs. They are the
central tables in the model. They represent important objects (or entities) within an organization. Hubs
don’t have many columns. There is always an identifier which is an artificial value having no meaning in
real life, it’s just there to represent a business object. It’s what’s sometimes called a surrogate key. In
addition, each hub contains a business key, a value that does have value to users. It’s important that this is
a real unique and mandatory value. Next, there are some metadata columns, such as the date when a
record was loaded in the Data Vault (META_LOAD_DTS). Note that this value may be different from the date
when the object represented by the record was really created. Also, the source of where the data comes
from is stored (META_SOURCE).

Satellites contain the attributes of the hubs. In this diagram H_CARRIER has two satellites, S_CARRIER and
S_CARRIER_USDOT_AIRLINE, and the hub H_AIRPORT has three, S_AIRPORT_OPENFLIGHTS, S_AIRPORT_RITA, and
S_AIRPORT_USDOT. Each hub table can have zero or more satellite tables. Each satellite contains the identifier
of the hub and the column that indicates the date when the satellite data was loaded. In fact, a record in a
satellite table represents a version of a hub’s satellite data. The column META_DATA_LOAD_DTS contains a date
that indicates when a record in the satellite has become obsolete. Most importantly a satellite table
contains a set of real attributes of a hub. This is where most of the data resides that users need for their
reporting. For example, the satellite S_AIRPORT_OPENFLIGHTS contains the attributes that indicate the
longitude and latitude coordinates of an airport.

This diagram only shows one link table called L_FLIGHTS. Usually, a Data Vault contains many links. A link
represents a relationship between two or more hubs. A link table has its own unique identifier, for each
relationship to a hub the identifier of the hub is included as foreign key. Because most link tables
represent an event, such as a money withdrawal from an ATM, a hospitalization, or a payment, they
usually contain one or more date and time columns indicating when the event took place. Like hubs, link
tables contain metadata columns, including the date on which the data is loaded in the Data Vault and the
source system where the data is copied from.

The L_FLIGHTS link table has two satellites, LSA_FLIGHT_KEYS_USA_DOMESTIC and LSA_FLIGHT_STATS_USA_DOMESTIC.
These satellite tables have the same purpose and the same structure as the satellites of the hubs except
that a satellite points to a link and not to a hub.

The role of the metadata columns, such as META_LOAD_DTS, META_LOAD_END_DTS, and META_SOURCE, is explained
in more detail in Section 4.

The Key Benefits of Data Vault – Other design and modeling techniques exist for developing data
warehouses. For example, classic normalization and star schema modeling5 are well‐known alternatives.
All have their strong and weak points. The two strong points of Data Vault are data model extensibility
and reproducibility:

 Data model extensibility refers to the ability to implement new information needs easily in the
Data Vault. In most data warehouses, if new information needs have to be implemented, existing
data structures must be changed. Those changes can lead to changes of ETL logic and to changes
of reports and applications while in fact these changes may not be relevant to them at all. With
Data Vault, new information leads to the introduction of new hubs, new satellites, or new links.

5
Wikipedia, Star Schema, see http://en.wikipedia.org/wiki/Star_schema, July 2014.

But never does it lead to changes of the existing data structures. They remain unaffected. So,
applications and reports for which the new information needs are irrelevant, don’t have to be
changed. Thus a change is easier to implement. In fact, the data model extensibility property of a
Data Vault makes it a highly flexible solution.

 Reproducibility is an aspect of compliance. Compliance is a state in which someone or something
is in accordance with established guidelines, specifications, or legislation. One aspect of
compliance is the need to reproduce reporting results or documents from the past. In a Data Vault
current and historical data is organized and stored in such a way that old and new data is always
available, and if the Data Vault approach is used, if a report has been created once, it can be
reproduced, even if the data models have changed. To support reproducibility all the metadata
columns are included in all the tables.

The Key Drawback of Data Vault – From the viewpoint of data model extensibility and reproducibility, Data
Vault has a lot to offer. But from a querying standpoint a Data Vault data model is complex. Because data
is distributed over many hubs, satellites, and links, developing reports straight on a Data Vault (as
indicated in Figure 2) leads to very complex SQL statements in which many tables must be joined in a
complex way. And complex queries are difficult to optimize by SQL database servers, often resulting in
poor query performance.

Figure 2 Running reports and queries straight on a data vault leads to complex queries and poor query performance.

For this reason, many organizations owning a Data Vault, have developed numerous derived data stores,
such as SQL‐based data marts, cubes for Microsoft SQL Server, spreadsheet files, and Microsoft Access
databases; see Figure 3. These derived data stores have data structures that are more suitable for
reporting and offer a much better performance.



Figure 3 To improve query performance and to present easier-to-understand data structures, many derived data stores are
developed.

Although such a solution solves the performance problems, it introduces some other ones:

 Refreshing derived data stores: All these derived data stores have to be designed, loaded with
data from the Data Vault, and they have to be managed. ETL programs have to be developed and
maintained to periodically refresh the data stores to keep the data up to date. They require a
scheduling environment. All this work is time‐consuming and costly.

 Data model changes: If a new satellite or link is added to a Data Vault, all the ETL programs
accessing those data structures have to be changed and all the derived data stores as well. Data
may have to be unloaded first, adapted ETL programs have to be executed to bring the derived
data store in synch, and finally data can be reloaded back in. Some technologies used to
implement derived data stores require a full recreate and full reload. All this does not improve the
time‐to‐market. Also, all these derived data stores can end up being a maintenance nightmare.

 No up‐to‐date data: When users query a derived data store, they don’t see 100% up‐to‐date data.
There is a big chance they see outdated data, because their derived data store is refreshed, for
example, only once every night.

 Redundant data storage: It’s obvious that this solution leads to redundant storage of data across
all these derived data stores.

To summarize, the intended flexibility that organizations try to achieve by implementing a Data Vault is
annulled because of the need to develop many derived data stores.


Data Virtualization – The challenge for many organizations using Data Vault is to develop an environment

that offers data model extensibility and reproducibility and that in addition allows for easy reporting and
management. Furthermore, this environment shouldn’t require the development of numerous derived
data stores. Such an environment can be developed using data virtualization. Here, a data virtualization
server replaces all the physical and derived data stores; see Figure 4. The reports see data structures that
are easy to access, easy to adapt, and they do contain up‐to‐date data.

Figure 4 The data virtualization server replaces all the derived data stores.

A Short History of Data Vault – Data Vault is not a new design approach. Its history dates back to 1990. In
2000 it became a public domain modeling method. Especially between 2005 and 2010 it became well‐
accepted. Currently, numerous organizations have adopted Data Vault and have used it to develop their
data warehouse. Because Data Vault offers a high level of reproducibility, many financial organizations
were quick to see its value. Currently, acceptance is spreading. A website6 exists that contains a list of
organizations who either use, build, or support Data Vault.

Dan Linstedt2 was the founding father and is still promoting Data Vault worldwide. Hans Hultgren has also
contributed much to evangelize data vault. His book3 on Data Vault definitely helped to explain Data Vault
to a large audience.

As has happened with many design approaches not managed by a standardization organization, Data
Vault is not always Data Vault. There are different styles. In this whitepaper one of the more common
approaches is selected. But if readers use a slightly different form, they will have no problem adapting the
approach to their own method and needs.

6
Dan Linstedt, Data vault Users, see http://danlinstedt.com/about/dv‐customers/#

3 The SuperNova Data Modeling Technique

Introduction – This section introduces the data modeling technique called SuperNova. It has been designed
to be used in conjunction with Data Vault and leans heavily on data virtualization technology. Step by step
it explains how to develop an environment that extracts data from a Data Vault and makes that data
available for reporting using a data virtualization server. The value of SuperNova is that the key benefits of
data vault are preserved: reproducibility and data model extensibility.

The result of applying the SuperNova modeling technique results in a set of tables. These tables are
implemented using the CIS data virtualization server. SuperNova works together with Data Vault and
extends it. The goal of a SuperNova is to make the data stored in a Data Vault easily and efficiently
accessible to a wide range of reports and analytics without jeopardizing reproducibility and data model
extensibility and without the need to develop and manage countless derived data stores. So, SuperNova is
a set of design and modeling techniques to, given an existing Data Vault, develop a layered set of data
models implemented with CIS.

The Name SuperNova – The SuperNova modeling technique is named after one of the layers in the solution.
It’s called SuperNova because when the development of this technique started, it was known that the
only workable solution would involve combining many hub, link, and satellite tables together, which
would lead to large and wide tables. So, the name had to represent something big, but the word big had
to be avoided, because that adjective is being overused today. Very quickly the term SuperNova was
coined. A supernova is massive. Some are over 10 light years across, and a supernova remnant is even
bigger: upwards of 50 million lights years. In short, something big was needed and a supernova was clearly
big enough. In addition, the word nova means “new” in Latin, and new the approach certainly is.

The SuperNova Architecture – Figure 5 represents the overall architecture of a SuperNova solution. The
diagram consists of five layers. The bottom layer in the diagram contains all the production databases. The
layer above represents a data warehouse designed using Data Vault. The top three layers together form
the SuperNova solution and are implemented using CIS.

Layer 1: Operational Systems – At the bottom of the diagram the operational systems are shown. These are
the systems in which new data is entered. These can be classic data entry systems, but also brand new
systems such as websites. These operational systems with their production databases fall outside the
control of the data virtualization server.

Layer 2: Data Vault EDW – Data is copied from the production databases to an enterprise data warehouse
designed using the Data Vault approach. This layer falls outside the control of the data virtualization
server as well. The copy process and the technology used are not determined by the data virtualization
server. In fact, this implementation is technologically irrelevant for the data virtualization server.
SuperNova assumes that the data warehouse has been designed according to the rules of the Data Vault
approach, that historical data is copied and stored in the Data Vault according to the rules, and so on.
Another assumption made is that the Data Vault model is a given. So, in the solution described no changes
are made or have to be made to that Data Vault. A SuperNova solution can be implemented on any
existing Data Vault.


Figure 5 A SuperNova solution consists of three layers, of which the top three are managed by the data virtualization server.

Layer 3: SuperNova Layer – Layer 3, the middle layer in this diagram, is the crux of the SuperNova solution.
It’s a set of views defined with CIS on the Data Vault tables. In these views, data from the Data Vault
tables is reorganized and restructured so that they are easy to access and that queries become
straightforward. The structure of these tables form a mixture of a Data Vault model and a star schema.
The tables are slightly denormalized and thus contain duplicate data.

The logic to extract data from the Data Vault tables to the SuperNova tables is relatively easy. The
structure of each SuperNova table is aimed at a wide range of reporting and analytical use cases. Basically,
satellite data is merged with hub data. A record in a SuperNova hub represents a version of a hub and it
contains all the satellite data relevant for that specific version. The link tables are also extended with their
own satellite data.

A SuperNova data model is extensible. For example, if a new satellite is added to a hub in the Data Vault
layer, the corresponding SuperNova hub must be redefined, but the existing reports won’t have to be
changed. Or, if new links between hubs are introduced, a new SuperNova link is created. If new hubs are
added, a new SuperNova hub is introduced. To summarize, the data model of the SuperNova layer is
almost as extensible as that of the data vault model.

Whether these views are materialized (cached) or not depends on many factors, including the query
workload, the database size, and the reporting requirements. This topic is discussed in Section 7.


Layer 4: Extended SuperNova Layer – The Extended SuperNova (XSN) layer consists of a set of views defined

on top of the SuperNova views using CIS. In this layer the views of the SuperNova layer are extended and
changed. For example, extra columns with derived data are added, column values are transformed and
standardized, data is filtered, and column names are standardized. In fact, business logic is added to raise
the quality level and the usefulness of the SuperNova model. The intention of this model is to contain
logic that is relevant to all users, not a subset. So, all the logic implemented in the XSN layer should apply
to all users and all reports.

Layer 5: Data Delivery Layer – The top layer is called the Data Delivery layer. This layer consists of views
accessed by users and reports. For each user group or reporting environment a set of views is defined that
fits their needs and works well with the technical requirements of the reporting and analytical tools. For
example, some reports require tables to be organized as a star schema, while others prefer wide, flat,
highly‐denormalized tables. Some of the users are only interested in current data, while others need
historical data as well. The structure and contents of the Data Delivery views are aimed at what the users
and reports need. In a way, some of these sets of views can be qualified as virtual data marts.

Overall Comments – The value of the SuperNova solution is that the key benefits of Data Vault are
preserved: reproducibility and data model extensibility. In addition, new derived data stores are very easy
to create and maintain, because they are completely virtual. No real data stores have to be developed and
managed. Although, if needed, the views of the SuperNova layer may have to be materialized for
performance or scalability reasons. So, the price is storage.

The layered model approach recommended with SuperNova aligns with common data virtualization
modeling best practices7.

4 Rules For the Data Vault Model

Designing and developing a Data Vault is a very rigid exercise. The modeling rules are very precise and
clear. Still, different developers have different opinions. Therefore, there can be slight variations. These
variations don’t have a major impact on the SuperNova modeling technique, but can lead to minor
alterations.

To avoid any confusion, here are the rules assumed in this whitepaper:

 Each hub table contains an identifier (ID) representing a hub object with a meaningless key. Each
should also contain the metadata columns BUSINESS_KEY, META_LOAD_DTS, and META_SOURCE. The first
one contains a meaningful key known by the users identifying the hub objects uniquely. The
second one indicates the date and time when a hub record was loaded in the Data Vault. The third
column indicates the source system (the bottom layer in Figure 5) from which the data was copied
to the Data Vault. There is no column indicating the enddate, because the assumption is made
that hub objects are never thrown away, not even when the business objects they represent don’t
exist anymore. This is required to offer reproducibility.

7
R. Eve, What to Do About the Data Silo Challenge, March 2013, see https://roberteve1.sys‐con.com/node/2591842

 Each satellite table contains the metadata columns HUB_ID, META_LOAD_DTS, META_LOAD_END_DTS, and
META_SOURCE. The first column contains the identifier of the hub to which the satellite belongs, the
second column indicates the date and time when a satellite record is loaded in the table, and the
third column indicates the date and time on which the record became obsolete. This is the date
on which a specific set of attributes has become obsolete. The fourth column, META_SOURCE,
indicates the source system from which the data is copied.

 Each hub table contains two special records. The first one identifies a so‐called Unknown hub
object. It has as identifier -1, and its business key is set to Unknown. The second record represents a
so‐called Not applicable hub record. Its identifier is set to -2 and its business key is set to N.a..
These two records are needed to link records in link tables to the hub tables even if there is no
(real) hub record for that link. For both records META_LOAD_DTS is set to January 1, 1900 and the
META_SOURCE is set to Repository.

 Each record in a satellite table identifies a version of a hub object. So, a specific hub object can
appear multiple times in a satellite table.

 When an enddate such as META_DATA_END_DTS is unknown, not the null value but the value
December 31, 9999 is inserted. This is the highest possible date in most SQL database servers. The
reason is twofold. First, many conditions used in the definitions of the SuperNova views make use
of the condition DATE BETWEEN STARTDATE AND ENDDATE. This condition works best if the end date has a
real value and not the null value. When null would be used, the condition would look like DATE
BETWEEN STARTDATE AND ENDDATE OR (DATE >= STARTDATE AND ENDDATE IS NULL). This condition is more
difficult to optimize, and thus may lead to poor query performance.

 Link tables can have eventdates. These are real values. These real business dates can fall before
the META_LOAD_DTS value.

 With respect to versions and historical data in the satellite tables, the following rules apply. Time‐
wise no gaps are allowed between two consecutive versions of a hub object. If a record does not
represent the last version, its enddate is equal to the startdate of the next version + 1. Here, 1 is
the smallest possible precision, such as a second or microsecond. Two versions of a hub in one
satellite are not allowed to overlap. At a particular point in time, only one version of a hub object
in a satellite table can exist. Two versions of a hub in two different satellites are allowed to
overlap.

 With respect to data cleansing, in the Data Vault approach it’s recommended not to change,
correct or cleanse the data when it’s copied from the production databases to the Data Vault. If
data has to be cleansed, it’s done after it has been loaded. In fact, the corrected data becomes the
new version of the data. The incorrect, old data remains in the Data Vault just in case there is a
report that was once executed on that incorrect data and must be reproduced.


5 The Sample Data Vault

In this whitepaper a small sample Data Vault is used to illustrate how SuperNova and its layers work. It
consists of three hubs, one link, and five satellite tables. The first hub has two satellites, the second one
only one, and the third none. The link table has two satellites. An overview of the data model of this set of
tables is presented in Figure 6.

Figure 6 The data model of the sample data vault.

Next, the contents of each table is shown. On purpose, the number of rows in each table is kept small to
be able to easily show what the results are of some of the steps described in the next sections.

HUB1
HUB_ID BUSINESS_KEY META_LOAD_DTS META_SOURCE
1 b1 2012-06-01 00:00:00 Demo
2 b2 2011-03-20 00:00:00 Demo
3 b3 2013-11-22 00:00:00 Demo
4 b4 2013-09-09 00:00:00 Demo
-1 Unknown 1900-01-01 00:00:00 Repository
-2 N.a. 1900-01-01 00:00:00 Repository


HUB1_SATELLITE1
HUB_ID META_LOAD_DTS META_LOAD_END_DTS ATTRIBUTE
1 2012-06-01 00:00:00 2013-11-14 23:59:59 a1
1 2013-11-15 00:00:00 2014-03-06 23:59:59 a2
1 2014-03-07 00:00:00 9999-12-31 00:00:00 a3
2 2011-03-20 00:00:00 2012-02-25 23:59:59 a4
2 2012-02-26 00:00:00 2014-02-25 23:59:59 a5
2 2014-02-26 00:00:00 9999-12-31 00:00:00 a6

HUB1_SATELLITE2
1 2013-06-21 00:00:00 2013-07-20 23:59:59 a7
1 2013-07-21 00:00:00 2013-11-12 23:59:59 a8
1 2013-11-13 00:00:00 9999-12-31 00:00:00 a9
2 2011—03-20 00:00:00 2012-02-25 23:59:59 a10
2 2012—02-26 00:00:00 9999-12-31 00:00:00 a11
3 2013—09-09 00:00:00 2013-11-11 23:59:59 a12
3 2013—11-12 00:00:00 2013-11-30 00:00:00 a13

HUB2
5 b5 2012-06-01 00:00:00 Demo
6 b6 2011-03-20 00:00:00 Demo
7 b7 2013-11-22 00:00:00 Demo
-2 N.a. 1900-01-01 00:00:00 Repository

HUB2_SATELLITE1
5 2011-11-09 00:00:00 2012-07-20 23:59:59 a14
5 2012-07-21 00:00:00 2014-03-21 23:59:59 a15
5 2014-03-22 00:00:00 9999-12-31 00:00:00 a16
6 2013-12-08 00:00:00 9999-12-31 00:00:00 a17

HUB3
8 b8 2013-05-15 00:00:00 Demo
9 b9 2013-09-20 00:00:00 Demo
-2 N.a. 1900-01-01 00:00:00 Repository

LINK
LINK_ID HUB1_ID HUB2_ID EVENTDATE META_LOAD_DTS META_SOURCE
1 1 5 2013-12-01 00:00:00 2013-12-02 00:00:00 Demo
2 1 6 2014-01-01 00:00:00 2014-01-01 00:00:00 Demo
3 2 6 2013-12-25 00:00:00 2013-12-26 00:00:00 Demo
4 3 -1 2013-06-24 00:00:00 2013-06-24 00:00:00 Demo


LINK_SATELLITE1
LINK_ID META_LOAD_DTS META_LOAD_END_DTS META_SOURCE ATTRIBUTE
1 2013-12-01 00:00:00 2014-01-23 23:59:59 Demo a18
1 2014-01-24 00:00:00 9999-12-31 00:00:00 Demo a19
3 2013-12-27 00:00:00 9999-12-31 00:00:00 Demo a20
4 2013-12-08 00:00:00 9999-12-31 00:00:00 Demo a21

LINK_SATELLITE2
LINK_ID META_LOAD_DTS META_LOAD_END_DTS META_SOURCE ATTRIBUTE
1 2013-12-25 00:00:00 9999-12-31 00:00:00 Demo a22
2 2014-03-12 00:00:00 9999-12-31 00:00:00 Demo a23
3 2014-02-02 00:00:00 9999-12-31 00:00:00 Demo a24

6 Developing the SuperNova Model

Introduction – A SuperNova model (the middle layer in Figure 5) is derived from a Data Vault model
according to a number of clear steps. A SuperNova model contains only link and hub tables. All the data of
the satellite tables is merged with their respective hub or link tables. In a SuperNova model a record in a
hub doesn’t represent a business object, but a version of that business object. The same applies to a
record in a link table. The model structure resembles that of a star schema in which fact tables are
represented by links and dimensions by hubs. This section describes these steps in detail using the sample
database described in Section 5. Figure 7 shows a model of the resulting SuperNova model after the
design steps have been applied.

Figure 7 The SuperNova data model derived from the
sample data vault.


Step 1 – Before the real SuperNova views can be defined, a layer of other views must be defined first.
These views contain temporary results. For each hub table in the data vault a view is created that contains
all the versions of each hub object. These versions are derived from the satellite tables of the hub. For
example, HUB1_SATELLITE1 contains the following three records for hub object 1:

HUB_ID META_LOAD_DTS META_LOAD_END_DTS
1 2012-06-01 00:00:00 2013-11-14 23:59:59
1 2013-11-15 00:00:00 2014-03-06 23:59:59
1 2014-03-07 00:00:00 9999-12-31 00:00:00

and HUB1_SATELLITE2 the following three records :

HUB_ID META_LOAD_DTS META_LOAD_END_DTS
1 2013-06-21 00:00:00 2013-07-20 23:59:59
1 2013-07-21 00:00:00 2013-11-12 23:59:59
1 2013-11-13 00:00:00 9999-12-31 00:00:00

These records do not really represent versions of hub object 1. The real versions become visible when the
two tables are merged together:

HUB_ID STARTDATE ENDDATE
1 2012-06-01 00:00:00 2013-06-20 23:59:59
1 2013-06-21 00:00:00 2013-07-20 23:59:59
1 2013-07-21 00:00:00 2013-11-12 23:59:59
1 2013-11-13 00:00:00 2013-11-14 23:59:59
1 2013-11-15 00:00:00 2014-03-06 23:59:59
1 2014-03-07 00:00:00 9999-12-31 00:00:00

In this result, two versions of the same hub object don’t overlap. A new version ends one second before
the start of a new version. If a satellite record of a hub object in HUB1_SATELLITE1 has the same startdate
and enddate as one in HUB1_SATELLITE2, they form one version. The current version of a hub object has
December 31, 9999 as enddate, unless an hub object is deleted. In this case, the delete date is entered as
the enddate.

The diagram in Figure 8 illustrates how data from multiple satellites is merged to find all the versions. The
top line indicates all the dates appearing in the startdate and enddate columns of the two satellite tables.
The second line indicates the dates on which new versions for hub 1 start in satellite 1, and the third line
indicates the new versions in satellite2. Underneath the dotted line the “sum” of the two is shown. It’s
clear that hub 1 has six versions. The first one starts at 2012-06-01, the second one at 2013-06-21, and so on.
Number six is the current version and therefore has no enddate. The enddate of each version is equal to
the startdate of the next version minus 1 second.



Figure 8 An illustration of all the versions of hub object 1.

The definition to create the view that contains all the versions of all objects of a hub:
CREATE VIEW HUB1_VERSIONS AS
WITH STARTDATES (HUB_ID, STARTDATE, ENDDATE, BUSINESS_KEY) AS (
SELECT HUB1.HUB_ID, SATELLITES.STARTDATE, SATELLITES.ENDDATE, HUB1.BUSINESS_KEY
FROM HUB1 LEFT OUTER JOIN
(SELECT HUB_ID, META_LOAD_DTS AS STARTDATE, META_LOAD_END_DTS AS ENDDATE
FROM HUB1_SATELLITE1
UNION
SELECT HUB_ID, META_LOAD_DTS, META_LOAD_END_DTS
FROM HUB1_SATELLITE2) AS SATELLITES
ON HUB1.HUB_ID = SATELLITES.HUB_ID)
SELECT DISTINCT HUB_ID, STARTDATE,
CASE WHEN ENDDATE_NEW <= ENDDATE_OLD THEN ENDDATE_NEW ELSE ENDDATE_OLD END AS ENDDATE,
BUSINESS_KEY
FROM (SELECT S1.HUB_ID, ISNULL(S1.STARTDATE,'1900-01-01 00:00:00') AS STARTDATE,
(SELECT ISNULL(MIN(STARTDATE - '1' SECOND),'9999-12-31 00:00:00')
FROM STARTDATES AS S2
WHERE S1.HUB_ID = S2.HUB_ID
AND S1.STARTDATE < S2.STARTDATE) AS ENDDATE_NEW,
ISNULL(S1.ENDDATE,'9999-12-31 00:00:00') AS ENDDATE_OLD,
S1.BUSINESS_KEY
FROM STARTDATES AS S1) AS S3

Note: This view applies the algorithm described in the previous text. In real systems, the greenish‐blue
terms must be replaced by the names of real tables and columns. The rest remains the same. This applies
to the other SQL statements in this whitepaper as well.

Explanation: In SQL, the query in the WITH clause is called a common table expression (CTE). This CTE
contain the following query:

SELECT HUB_ID, META_LOAD_DTS AS STARTDATE, META_LOAD_END_DTS AS ENDDATE
UNION
SELECT HUB_ID, META_LOAD_DTS, META_LOAD_END_DTS


For each satellite of the hub the values of the identifier, the startdate and enddate are selected. These

results are combined using a UNION operator. The temporary result of this query is called SATELLITES. It
contains the startdates and enddates off the records of all the satellites of a hub. If for a hub object
multiple satellite records exist with the same startdate and enddate, the all except one are removed by
the UNION operator. The (intermediate) result of this query is:

SATELLITES
HUB_ID STARTDATE ENDDATE
1 2012-06-01 00:00:00 2013-11-14 23:59:59
1 2013-06-21 00:00:00 2013-07-20 23:59:59
1 2013-07-21 00:00:00 2013-11-12 23:59:59
1 2013-11-13 00:00:00 9999-12-31 00:00:00
1 2013-11-15 00:00:00 2014-03-06 23:59:59
1 2014-03-07 00:00:00 9999-12-31 00:00:00
2 2011-03-20 00:00:00 2012-02-25 23:59:59
2 2012-02-26 00:00:00 2014-02-25 23:59:59
2 2012-02-26 00:00:00 9999-12-31 00:00:00
2 2014-02-26 00:00:00 9999-12-31 00:00:00
3 2013-09-09 00:00:00 2013-11-11 00:00:00
3 2013-11-12 00:00:00 2013-11-12 00:00:00

Note that this result does not include hub object 4, because it has no satellite data.

If new satellite tables are added to a hub table, only the CTE has to be changed. It has to be extended with
an extra subquery that is union‐ed with the others.

In this view definition the META_LOAD_DTS and META_LOAD_END_DTS columns are used as representatives for the
startdates and enddates. It could be that a satellite table also contains other types of startdates and
enddates, such as the transaction dates (the dates on which the data was inserted in the IT system), or the
valid dates (the dates on which the event really happened in real life). The information needs determine
which set of dates is selected. There is no impact on the rest of the algorithm.

Next, the CTE is executed. Here, the original hub table itself is joined with this intermediate result called
SATELLITES. The result of this CTE is called STARTDATES. A left outer join is used to guarantee that even hub
objects with no satellites are included. For such objects, the startdate and enddate are set to null. The
business key of the hub is included in this result as well. In fact, each column from the hub table that
needs to be included in the SuperNova views, is added here. The contents of STARTDATES is:

STARTDATES
HUB_ID STARTDATE ENDDATE BUSINESS_KEY
1 2012-06-01 00:00:00 2013-06-20 23:59:59 b1
1 2013-06-21 00:00:00 2013-07-20 23:59:59 b1
1 2013-07-21 00:00:00 2013-11-12 23:59:59 b1
1 2013-11-13 00:00:00 2013-11-14 23:59:59 b1
1 2013-11-15 00:00:00 2014-03-06 23:59:59 b1
1 2014-03-07 00:00:00 9999-12-31 00:00:00 b1
2 2011-03-20 00:00:00 2012-02-25 23:59:59 b2
2 2012-02-26 00:00:00 2014-02-25 23:59:59 b2
2 2012-02-26 00:00:00 9999-12-31 00:00:00 b2
2 2014-02-26 00:00:00 9999-12-31 00:00:00 b2
3 2013-09-09 00:00:00 2013-11-11 00:00:00 b3 Table continues on the next page

3 2013-11-12 00:00:00 2013-11-12 00:00:00 b3

4 NULL NULL b4
-1 NULL NULL Unknown
-2 NULL NULL N.a.

There are three rows with no startdate and no enddate. The first one doesn’t have them, because hub
object 4 has no satellites. The other two rows have identifiers less than 0. These rows do not represent
real hub objects, but are needed for the link tables; see Section 4.

This is the result of the main query of the view definition, or in other words, this is the virtual contents of
the view called HUB1_VERSIONS:

HUB1_VERSIONS
HUB_ID STARTDATE ENDDATE HUB_BUSINESS_KEY
1 2012-06-01 00:00:00 2013-06-20 23:59:59 b1
1 2013-06-21 00:00:00 2013-07-20 23:59:59 b1
1 2013-07-21 00:00:00 2013-11-12 23:59:59 b1
1 2013-11-13 00:00:00 2013-11-14 23:59:59 b1
1 2013-11-15 00:00:00 2014-03-06 23:59:59 b1
1 2014-03-07 00:00:00 9999-12-31 00:00:00 b1
2 2011-03-20 00:00:00 2012-02-25 23:59:59 b2
2 2102-02-26 00:00:00 2014-02-25 23:59:59 b2
2 2014-02-26 00:00:00 9999-12-31 00:00:00 b2
3 2013-09-09 00:00:00 2013-11-11 23:59:59 b3
3 2013-11-12 00:00:00 2013-11-30 00:00:00 b3
4 1900-01-01 00:00:00 9999-12-31 00:00:00 b4
-1 1900-01-01 00:00:00 9999-12-31 00:00:00 Unknown
-2 1900-01-01 00:00:00 9999-12-31 00:00:00 N.a.

This table consists of the following four columns:

 Hub identifier: This column is copied from the STARTDATES result.
 Startdate: This value is copied from the STARTDATES result unless it is null, then the start date is set
to January 1, 1900. This date represents the lowest possible date. It means that the start date of
this hub version is not really known.
 Enddate: The end date is determined with a subquery. For each record the subquery retrieves the
startdates of all the records belonging to the same hub object and determines what the oldest is
of all the ones not older than the startdate of the record itself. Next, 1 second is subtracted from
that startdate and that value is seen as the enddate of that version. If the startdate is null, as is
the case for hubs without satellites, this version is the last version and the enddate is set to
December 31, 9999. This makes it the current version for that hub object.
 Business key: The business key is copied from the STARTDATES result.

The above view definition is required if a hub has two or more satellites. If a hub has no satellites, the
definition of its VERSIONS view looks much simpler:

CREATE VIEW HUB3_VERSIONS (HUB_ID, STARTDATE, ENDDATE, BUSINESS_KEY) AS
SELECT HUB_ID, ISNULL(META_LOAD_DTS, '1900-01-01 00:00:00'),
'9999-12-31 00:00:00', BUSINESS_KEY
FROM HUB3


Because there are no satellites, no values for startdate end enddate exist. Therefore, they are respectively
set to January 1, 1900 and December 31, 9999. This means that for each hub object one version is
created, and that version is the current one.

This same view can easily be defined using the Grid of CIS; see Figure 9.

Figure 9 Defining the HUB3_VERSIONS view using the Grid module of CIS.

When a hub has one satellite, such as is the case with the HUB2 table, this is the preferred view definition:

CREATE VIEW HUB2_VERSIONS (HUB_ID, STARTDATE, ENDDATE, BUSINESS_KEY) AS
SELECT HUB2.HUB_ID, ISNULL(HUB2_SATELLITE1.META_LOAD_DTS, '1900-01-01 00:00:00'),
ISNULL(HUB2_SATELLITE1.META_LOAD_END_DTS, '9999-12-31 00:00:00'),
HUB2.BUSINESS_KEY
FROM HUB2 LEFT OUTER JOIN HUB2_SATELLITE1
ON HUB2.HUB_ID = HUB2_SATELLITE1.HUB_ID

The META_LOAD_DTS stored in the hub table can be used as the startdate, and the enddate is set to
December 31, 9999. If the hub does contain some enddate, it’s recommended to use that one.

Step 2 – Now that for each hub table in the Data Vault Model a view has been defined that shows all the
versions of each hub object, the SuperNova version of each hub table can be defined. The SuperNova
version contains a record for each version of a hub object plus all the attributes coming from all the
satellite tables. With respect to the metadata columns, they are all left out, because they don’t represent
information needs.


View definition:

CREATE VIEW SUPERNOVA_HUB1
(HUB_ID, STARTDATE, ENDDATE, BUSINESS_KEY, ATTRIBUTE1, ATTRIBUTE2) AS
SELECT HUB1_VERSIONS.HUB_ID, HUB1_VERSIONS.STARTDATE, HUB1_VERSIONS.ENDDATE,
HUB1_VERSIONS.BUSINESS_KEY,
HUB1_SATELLITE1.ATTRIBUTE, HUB1_SATELLITE2.ATTRIBUTE
FROM HUB1_VERSIONS
LEFT OUTER JOIN HUB1_SATELLITE1
ON HUB1_VERSIONS.HUB_ID = HUB1_SATELLITE1.HUB_ID AND
(HUB1_VERSIONS.STARTDATE <= HUB1_SATELLITE1.META_LOAD_END_DTS
AND HUB1_VERSIONS.ENDDATE >= HUB1_SATELLITE1.META_LOAD_DTS)

The contents of this SUPERNOVA_HUB1 view:

SUPERNOVA_HUB1
HUB_ID STARTDATE ENDDATE HUB_BUSINESS ATTRIBUTE1 ATTRIBUTE2
_KEY
1 2012-06-01 00:00:00 2013-06-20 23:59:59 b1 a1 NULL
1 2013-06-21 00:00:00 2013-07-20 23:59:59 b1 a1 a7
1 2013-07-21 00:00:00 2013-11-12 23:59:59 b1 a1 a8
1 2013-11-13 00:00:00 2013-11-14 23:59:59 b1 a1 a9
1 2013-11-15 00:00:00 2014-03-06 23:59:59 b1 a2 a9
1 2014-03-07 00:00:00 9999-12-31 00:00:00 b1 a3 a9
2 2011-03-20 00:00:00 2012-02-25 23:59:59 b2 a4 a10
2 2102-02-26 00:00:00 2014-02-25 23:59:59 b2 a5 a11
2 2014-02-26 00:00:00 9999-12-31 00:00:00 b2 a6 a11
3 2013-09-09 00:00:00 2013-11-11 23:59:59 b3 NULL a12
3 2013-11-12 00:00:00 2013-11-30 00:00:00 b3 NULL a13
4 1900-01-01 00:00:00 9999-12-31 00:00:00 b4 NULL NULL
-1 1900-01-01 00:00:00 9999-12-31 00:00:00 Unknown NULL NULL
-2 1900-01-01 00:00:00 9999-12-31 00:00:00 N.a. NULL NULL

In this view the HUB_VERSIONS view is joined with each satellite table. The complex part of this view
definition is formed by the used join conditions. Here is the join condition to join HUB1 with HUB_SATELLITE1:

HUB1_VERSIONS.HUB_ID = HUB1_SATELLITE1.HUB_ID AND

This join condition determines to which version in the HUB_VERSIONS table a hub relates in the
HUB1_SATELLITE1 table. The trick is to determine whether the startdate‐enddate pair of the hub version
overlaps with a satellite’s startdate‐enddate pair. This is the case when the startdate of the hub falls
before the satellite’s enddate and if the enddate of the hub falls after the satellite’s startdate.

Whether all the attributes of a satellite should be included, depends on the information needs. Adding
some later is a relatively simple exercise, because only the view definition has to be changed.


For the HUB2 table that has only one satellite table, the view definition is almost identical, except there is
only one left outer join:

(HUB_ID, STARTDATE, ENDDATE, BUSINESS_KEY, ATTRIBUTE1) AS
SELECT HUB2_VERSIONS.HUB_ID, HUB2_VERSIONS.STARTDATE, HUB2_VERSIONS.ENDDATE,
HUB2_VERSIONS.BUSINESS_KEY, HUB2_SATELLITE1.ATTRIBUTE
FROM HUB2_VERSIONS

For hubs without satellites, the view definition is very straightforward, because there are no attributes
and thus no joins:

(HUB_ID, STARTDATE, ENDDATE, BUSINESS_KEY) AS
SELECT HUB_ID, STARTDATE, ENDDATE, BUSINESS_KEY
FROM HUB3_VERSIONS

Step 3 – For each link table we must go through the same process as for the hubs. For each link table in
the Data Vault a view is created that contains all the versions of each link object.

CREATE VIEW LINK_VERSIONS AS
WITH STARTDATES (LINK_ID, STARTDATE, ENDDATE, HUB1_ID, HUB2_ID, EVENTDATE) AS (
SELECT LINK.LINK_ID, SATELLITES.STARTDATE, SATELLITES.ENDDATE,
LINK.HUB1_ID, LINK.HUB2_ID, LINK.EVENTDATE
FROM LINK LEFT OUTER JOIN
(SELECT LINK_ID, META_LOAD_DTS AS STARTDATE, META_LOAD_END_DTS AS ENDDATE
FROM LINK_SATELLITE1
UNION
SELECT LINK_ID, META_LOAD_DTS, META_LOAD_END_DTS
FROM LINK_SATELLITE2) AS SATELLITES
ON LINK.LINK_ID = SATELLITES.LINK_ID)
SELECT DISTINCT LINK_ID, STARTDATE,
CASE WHEN ENDDATE_NEW <= ENDDATE_OLD THEN ENDDATE_NEW ELSE ENDDATE_OLD END AS ENDDATE,
HUB1_ID, HUB2_ID, EVENTDATE
FROM (SELECT S1.LINK_ID, ISNULL(S1.STARTDATE, '1900-01-01') AS STARTDATE,
(SELECT ISNULL(MIN(STARTDATE - INTERVAL '1' SECOND),'9999-12-31 00:00:00')
FROM STARTDATES AS S2
WHERE S1.LINK_ID = S2.LINK_ID
AND S1.STARTDATE < S2.STARTDATE) AS ENDDATE_NEW,
ISNULL(S1.ENDDATE,'9999-12-31') AS ENDDATE_OLD,
S1.HUB1_ID, S1.HUB2_ID, S1.EVENTDATE
FROM STARTDATES AS S1) AS S3


Contents of the SuperNova view for the link table:

LINK_VERSIONS
LINK_ID STARTDATE ENDDATE HUB1_ID HUB2_ID EVENTDATE
1 2013-12-01 00:00:00 2013-12-24 23:59:59 1 5 2013-12-01
1 2013-12-25 00:00:00 2014-01-23 23:59:59 1 5 2013-12-01
1 2014-01-24 00:00:00 9999-12-31 00:00:00 1 5 2013-12-01
2 2014-03-12 00:00:00 9999-12-31 00:00:00 1 6 2014-01-01
3 2013-12-27 00:00:00 2014-02-01 23:59:59 2 6 2013-12-25
3 2014-02-02 00:00:00 9999-12-31 00:00:00 2 6 2013-12-25
4 2013-12-08 00:00:00 9999-12-31 00:00:00 3 -1 2013-06-24

Explanation: This view contains for each version of a link object a version. For each link version it shows
the startdate, enddate, the eventdate, and the identifiers of the hubs to which the link version points. In
this example, link 4 is linked to hub ‐1, which is the Unknown hub. In other words, there is no link to the HUB2
table.

Step 4 – As for the hubs, for each link table a SuperNova version must be defined that shows all the
versions for each link object:

CREATE VIEW SUPERNOVA_LINK
(LINK_ID, HUB1_ID, HUB2_ID, STARTDATE, ENDDATE, EVENTDATE,
ATTRIBUTE1, ATTRIBUTE2) AS
SELECT LINK_VERSIONS.LINK_ID, LINK_VERSIONS.HUB1_ID, LINK_VERSIONS.HUB2_ID,
LINK_VERSIONS.STARTDATE, LINK_VERSIONS.ENDDATE,
LINK_VERSIONS.EVENTDATE,
LINK_SATELLITE1.ATTRIBUTE, LINK_SATELLITE2.ATTRIBUTE
FROM LINK_VERSIONS
LEFT OUTER JOIN LINK_SATELLITE1
ON LINK_VERSIONS.LINK_ID = LINK_SATELLITE1.LINK_ID AND
(LINK_VERSIONS.STARTDATE <= LINK_SATELLITE1.META_LOAD_END_DTS
AND LINK_VERSIONS.ENDDATE >= LINK_SATELLITE1.META_LOAD_DTS)
LEFT OUTER JOIN LINK_SATELLITE2
ON LINK_VERSIONS.LINK_ID = LINK_SATELLITE2.LINK_ID AND
(LINK_VERSIONS.STARTDATE <= LINK_SATELLITE2.META_LOAD_END_DTS
AND LINK_VERSIONS.ENDDATE >= LINK_SATELLITE2.META_LOAD_DTS)

Contents of the SuperNova link view (the time component is not shown in the columns STARTDATE and
EVENTDATE, because they are all equal to 00:00:00):

SUPERNOVA_LINK
LINK_ID HUB1_ID HUB2_ID STARTDATE ENDDATE EVENTDATE ATTRIBUTE1 ATTRIBUTE2
1 1 5 2013-12-01 2013-12-24 23:59:59 2013-12-01 a18 NULL
1 1 5 2013-12-25 2014-01-23 23:59:59 2013-12-01 a18 a22
1 1 5 2014-01-24 9999-12-31 00:00:00 2013-12-01 a19 a22
2 1 6 2014-03-12 9999-12-31 00:00:00 2014-01-01 NULL a23
3 2 6 2013-12-27 2014-02-01 23:59:59 2013-12-25 a20 NULL
3 2 6 2014-02-02 9999-12-31 00:00:00 2013-12-25 a20 a24
4 3 -1 2013-12-08 9999-12-31 00:00:00 2013-06-24 a21 NULL

Step 5 – To make the views even easier to access, an alternative, “wider” version of the SuperNova link
tables can be defined. This SuperNova link table contains the identifiers and the startdates for each hub
that it is connected to. Those two columns identify a unique hub version. Note that this only works when

each link has some kind of eventdate: the date on which the link became relevant. So, this can be the
departure date of a flight or the date of a purchasing transaction.
CREATE VIEW SUPERNOVA_LINK_WIDE
(LINK_ID, LINK_STARTDATE, LINK_ENDDATE, EVENTDATE,
HUB1_ID, HUB1_STARTDATE, HUB2_ID, HUB2_STARTDATE,
ATTRIBUTE1, ATTRIBUTE2) AS
SELECT SUPERNOVA_LINK.LINK_ID,
SUPERNOVA_LINK.STARTDATE, SUPERNOVA_LINK.ENDDATE, SUPERNOVA_LINK.EVENTDATE,
ISNULL(SUPERNOVA_HUB1.HUB_ID, -1),
ISNULL(SUPERNOVA_HUB1.STARTDATE, '1900-01-01 00:00:00'),
ISNULL(SUPERNOVA_HUB2.HUB_ID, -1),
ISNULL(SUPERNOVA_HUB2.STARTDATE, '1900-01-01 00:00:00'),
SUPERNOVA_LINK.ATTRIBUTE1, SUPERNOVA_LINK.ATTRIBUTE2
FROM SUPERNOVA_LINK
LEFT OUTER JOIN SUPERNOVA_HUB1
ON SUPERNOVA_LINK.HUB1_ID = SUPERNOVA_HUB1.HUB_ID
AND (SUPERNOVA_LINK.EVENTDATE BETWEEN
SUPERNOVA_HUB1.STARTDATE AND SUPERNOVA_HUB1.ENDDATE)
SUPERNOVA_HUB2.STARTDATE AND SUPERNOVA_HUB2.ENDDATE)

Contents of the wide SuperNova link view:

SUPERNOVA_LINK_WIDE
LINK LINK_ LINK_ EVENTDATE HUB1 HUB1_ HUB2 HUB2_ ATTRIBUTE1 ATTRIBUTE2
_ID STARTDATE ENDDATE _ID STARTDATE _ID STARTDATE
1 2013-12-01 2013-12-24 2013-12-01 1 2013-11-15 5 2012-07-21 a18 NULL
23:59:59
1 2013-12-25 2014-01-23 2013-12-01 1 2013-11-15 5 2012-07-21 a18 a22
23:59:59
1 2014-01-24 9999-12-31 2013-12-01 1 2013-11-15 5 2012-07-21 a19 a22
00:00:00
2 2014-03-12 9999-12-31 2014-01-01 1 2013-11-15 6 2013-12-08 NULL a23
00:00:00
3 2013-12-27 2014-02-01 2013-12-25 2 2012-02-26 6 2013-12-08 a20 NULL
23:59:59
3 2014-02-02 9999-12-31 2013-12-25 2 2012-02-26 6 2013-12-08 a20 a24
00:00:00
4 2013-12-08 9999-12-31 2013-06-24 -1 1900-01-01 -1 1900-01-01 a21 NULL
00:00:00

Explanation: The SuperNova link table is joined with all its satellites. The trick here is to select the right
hub version for each link object. This is done by locating the link version whose eventdate falls between
the startdate and enddate of the hub version. If the eventdate of the link falls before the first version of
the related hub, the hub called Unknown (with value -1) is returned, and the startdate is set to January 1,
1900. If a link object has no hub record (because the relationship is optional or not relevant), it takes the
hub objects with values -1 or -2, which have the lowest possible startdate and the highest possible
enddate.

There are alternative ways to deal with links whose eventdate fall before the first version of a hub. It may
be useful to connect that link to the first version of the hub. In this case, the following join is required. The
join condition is extended with a subquery that selects the first version of the hub:


SUPERNOVA_HUB1.STARTDATE AND SUPERNOVA_HUB1.ENDDATE) OR
SUPERNOVA_LINK.EVENTDATE <
(SELECT MIN(SUPERNOVA_HUB1.STARTDATE)
FROM SUPERNOVA_HUB1
WHERE SUPERNOVA_LINK.HUB1_ID = SUPERNOVA_HUB1.HUB_ID)

There are two advantages of using wide views. First, because the wide version contains many columns,
users and reports will have to do less joins. This can have a positive impact on query performance.
Second, it allows for foreign keys to be defined between the link views and its hubs, which is required by
some reporting tools.

As an overview, Figure 10 presents all the other views and tables on which the SUPERNOVA_LINK_WIDE view is
dependent.

Figure 10 An overview of views and tables on which the SUPERNOVA_LINK_WIDE view is dependent.

Step 6 – Define primary keys on the SuperNova hub and link views. Because they are views, they don’t do
anything. Nevertheless, it’s recommended to define them, because many reporting and analytical tools
need to know what the primary key of each table is. By defining these keys, that metadata is exposed and
can be retrieved by the tools. The tools will understand the structure of the hubs and links better and they
can more intelligently generate queries. Figure 11 shows how the primary key for the SUPERNOVA_HUB1 view
is defined.


Figure 11 Definition in CIS of the primary key of the SUPERNOVA_HUB1 view.

Step 7 – What applies to primary keys, applies to foreign keys as well. Again, the reason is that many
reporting and analytical need to know the relationships (foreign keys) between the views. As example,
here is the SQL code to define two foreign keys from the SUPERNOVA_LINK_WIDE view to the two hub views.

ALTER TABLE SUPERNOVA_LINK_WIDE
ADD FOREIGN KEY (HUB1_ID, HUB1_STARTDATE) REFERENCES
SUPERNOVA_HUB1 (HUB_ID, STARTDATE)

ALTER TABLE SUPERNOVA_LINK_WIDE
ADD FOREIGN KEY (HUB2_ID, HUB2_STARTDATE) REFERENCES
SUPERNOVA_HUB2 (HUB_ID, STARTDATE)

Figure 12 shows how the foreign key linking the SUPERNOVA_LINK_WIDE view to the SUPERNOVA_HUB1 view is
defined.

Note that foreign keys cannot be defined for the original SuperNova link tables. The reason is that the
primary key of each hub consists of the hub identifier and the startdate. And it’s the startdate that’s
missing from the link table.

7 Optimizing the SuperNova Model

The performance of queries on the views of the SuperNova layer is dependent on many aspects, including
the hardware being deployed, the database server in use, whether caching is switched on, how much data
the views contain, and the query workload. This section contains some recommendations for optimizing
the SuperNova model when performance problems are encountered. Note that these are very generic
recommendations.



Figure 12 Define in CIS what the foreign key of a SuperNova view is.

Updating Statistical Information Of Data Vault Tables – A classic reason why performance may be poor is that
the statistical information on the tables and indexes in the Data Vault are out of date. They don’t reflect
the correct situation anymore. Because the query optimizer of a SQL database server bases its processing
strategy on that statistical information, it’s crucial that it’s always up to date. How statistical information
can be updated is product‐dependent. We refer to the manuals.

Caching of SuperNova Views – For those SuperNova views (links and hubs) that exhibit a poor query
performance, define a cache. Figure 13 contains a screenshot that indicates how a cache is defined on the
view SUPERNOVA_HUB1.

Figure 13 Define a cache for a SuperNova view.

It’s recommended that these caches are stored in a SQL database server. This has as an advantage that
extra indexes can be defined which speed up performance even more.

For each cached view, a refresh schedule must be defined. An example of a simple refresh schedule is:
every 24 hours at night using a full refresh. The refresh schedule depends on the requirements concerning
the actuality of data. With respect to the order of refreshes, first refresh all the SuperNova hub views and
then the SuperNova link views.

If performance on a view is still unacceptable, for example because it contains millions of rows and the
queries continuously accesses a large portion of the rows, it’s recommended to create a cache for the
VERSIONS view on which the view itself is defined. If that still not sufficient, it’s worth trying to turn the
STARTDATES CTE into a cached view as well.

Updating Statistical Information Of Caches – As it’s important that statistical information for Data Vault tables
is up to date, it’s also important that statistical information on the cached views is also up to date. Be sure
that after every refresh of a cache, an update of the statistical information is considered. Out‐of‐date
statistical information can lead to poor query performance.

Indexes on Cached Views – By default, CIS implements cached views with files. Access to those files can be
fast, but it’s not the fastest data storage technology available, especially if files become large. Therefore,
CIS allows caches to be stored in SQL database servers.

The advantage of using a SQL database server is that all its performance improving features can be
deployed to speed up access to the caches, such as memory buffer optimization, table partitioning, and
tablespace organization. One of those features is that extra indexes can be defined on the caches to speed
up query performance.

Defining indexes on the foreign keys in the SuperNova link tables makes sense. Also, define indexes on
columns that are aggregated often. The same applies for columns on which many filters are applied. Note
that adding extra indexes can slow down refreshing of the caches somewhat.

Developing the indexes on the caches can’t be done from within CIS. This must be done using the SQL
database server itself. When an extra index is defined, be sure that the right parameters are set; see the
manuals of the SQL database server.

Note: The structure of a table used for caching a view is a little different from the view itself. CIS adds one
column called CACHEKEY for internal processing. It contains an identifier of the refresh run. A separate table
indicates what that identifier means. Never change anything with respect to this column. Nor, should the
indexes that are defined by CIS itself be dropped.

Optimizing the SQL Database Server – Each SQL database server offers a wide range of features to improve
performance. Tables can be partitioned, buffers can be enlarged and tuned, table space parameters can
be optimized, data can be moved to faster storage technology, and so on. Check with database
administrators to see if they can tune and optimize access to the cached views.

Different Database Server – CIS allows caches to be stored in SQL database servers, such as EMC Greenplum,
HP Vertica, IBM DB2 LUW, IBM PureData for Analytics (Netezza), Microsoft SQL Server, MySQL, Oracle,
PostgreSQL, SAP Sybase ASE, Sybase IQ, and Teradata. Most of these are much more scalable and high‐

performing than files and will definitely speed up the queries on cached views.

8 Developing the Extended SuperNova Model

The views defined in the SuperNova layer present the data stored in the Data Vault but with a different
structure, one that is easier to use. But the layer doesn’t change the data nor does it filter out data, except
the metadata is left out. It’s primarily a re‐arrangement of the data. In the next layer, called the Extended
SuperNova layer (XSN), extra columns are added and values are transformed. This XSN layer consists of a
set of views defined on top of the SuperNova views. Each view in the XSN layer relates to one view in the
SuperNova layer (and vice versa). All the columns of a SuperNova view are included in the corresponding
XSN view. What’s different is that these XSN views can be extended with columns that contain derived
data and values may be transformed.

Adding Derived Data – An XSN view may include columns with derived data. For example, when birth dates
are available in a SuperNova hub for employees and users want to see the employee’s age as well, an
extra column can be defined that shows that age. Or, when a SuperNova view shows temperatures in
Fahrenheit, the XSN view contains the temperature in Celsius as well.

As an example, the next view is defined on the SUPERNOVA_HUB1 view and contains two derived columns. The
first one shows the number of days between the STARTDATE and the ENDDATE columns, and the second one
the date of the Sunday following the STARTDATE column.
CREATE VIEW XSN_HUB1 AS
SELECT HUB_ID, STARTDATE, ENDDATE, BUSINESS_KEY, ATTRIBUTE1, ATTRIBUTE2,
CASE WHEN STARTDATE <> '1900-01-01 00:00:00' AND
ENDDATE <> '9999-12-31 00:00:00'
THEN CAST(DAYS_BETWEEN(ENDDATE, STARTDATE) AS INTEGER)
ELSE NULL END AS DAYS_BETWEEN,
CASE WHEN STARTDATE <> '9999-12-31 00:00:00'
THEN CASE DAYOFWEEK(STARTDATE)
WHEN 1 THEN STARTDATE + INTERVAL '7' DAY
END
ELSE STARTDATE END AS NEXT_SUNDAY
FROM SUPERNOVA_HUB1

Explanation: The formula to derive the number of days in between is a little bit more complex than
expected. The reason is that we must cater for the situation that the startdate is January 1, 1900 and the
enddate is December 31, 9999. In both cases, the number of days in between is set to null. The formula
for determining the next Sunday after a particular data is also a little tricky.


Contents of the XSN_HUB1 view:

XSN1_HUB1
HUB_ID STARTDATE ENDDATE HUB_BUSINESS ATTRIBUTE1 ATTRIBUTE2 DAYS_ NEXT_
_KEY BETWEEN SUNDAY
1 2012-06-01 2013-07-20 23:59:59 b1 a1 a7 NULL 1900-01-07
1 2013-07-21 2013-11-12 23:59:59 b1 a1 a8 NULL 1900-01-07
1 2013-11-13 2013-11-14 23:59:59 b1 a1 a9 415 2012-06-03
1 2013-11-15 2014-03-06 23:59:59 b1 a2 a9 115 2013-07-28
1 2014-03-07 9999-12-31 00:00:00 b1 a3 a9 2 2013-11-17
2 2011-03-20 2012-02-25 23:59:59 b2 a4 a10 112 2013-11-17
2 2102-02-26 2014-02-25 23:59:59 b2 a5 a11 NULL 2014-03-09
2 2014-02-26 9999-12-31 00:00:00 b2 a6 a11 343 2011-03-27
3 2013-09-09 2013-11-11 23:59:59 b3 NULL a12 731 2012-03-04
3 2013-11-12 2013-11-30 00:00:00 b3 NULL a13 NULL 2014-03-02
4 1900-01-01 9999-12-31 00:00:00 b4 NULL NULL 64 2013-09-15
-1 1900-01-01 9999-12-31 00:00:00 Unknown NULL NULL 18 2013-11-17
-2 1900-01-01 9999-12-31 00:00:00 N.a. NULL NULL NULL 1900-01-07

Note: Include derived columns for which the formula requires only access to values stored in the same
row. So, don’t include aggregated data, move those to the next layer, the Data Delivery layer.

Transforming Data – The second purpose of the XSN views is transforming data. The definition of an XSN
view can include logic to transform values. For example, a column containing employee genders may
contain different codes indicating female, in the view definition these codes may be transformed to the
correct code. Or, if two values in a Data Vault column have been concatenated as one, it may be necessary
to split them as two separate values/columns. It may also be necessary to transform certain columns in
such a way that it becomes easier for users to integrate the data from multiple tables.

To develop more complex transformations, it may be necessary to define extra tables that, for example,
contain lists of incorrect values and their correct counterparts. These tables can be defined in the same
database where the views are cached, so that a join of the cached views and code tables can be pushed
down by CIS to that database server.

Reuse of Definitions – Defining XSN views has several practical advantages. First, by defining derived
columns and by transforming values at this layer, all the users share the same definitions regardless of the
reporting tool they use. This leads to more consistent reporting results. Second, it saves development
time. The logic has to be developed only once. Third, some of the simpler reporting tools don’t offer the
functionality to define derived columns or do complex transformation, which makes this a useful
alternative.

Always XSN – Even if there are no extra columns and even if there are no transformations needed, it’s
strongly recommended to define XSN views on top of the SuperNova views. If the XSN views exist, it
becomes easier to hide changes to the SuperNova views that are irrelevant for certain users or reports. In
other words, if the Data Vault tables change and the SuperNova views change accordingly, those changes
can be hidden for those users and reports not interested in the new data.


9 Developing the Data Delivery Model

The top layer in Figure 5 is called the Data Delivery layer (DD). DD views are defined on top of the XSN
views. For each user group a separate set of DD views is defined; see Figure 14. DD views show the data in
the form the users and reports need, or the form which it is most convenient for them. With respect to
their virtual contents DD views can overlap.

Figure 14 User groups have different sets of DD views.

As examples, five different forms of DD views are described here:

 Data is shown in a filtered manner
 Data is shown in aggregated form
 Data is shown in one large, highly denormalized table
 Data is shown in a star schema form
 Data is shown with a service interface

A Filtered View – A particular group of users may not be interested in all the data of an XSN view, or they
may not be allowed to access all the data of a view. For example, they may only want to work with sales
data for a particular sales region, or they may only see current data and not historical data. In such cases,
filters are added to the DD views defined on the XSN views.

An Aggregated View – Especially link tables may contain massive amounts of records showing data on a very
high level of detail. It may be more useful for users to aggregate the data somewhat. For example,
purchases may be aggregated per hour, and deliveries per week or per product group. In this case, group
by‐operators are added to the DD views defined on the XSN views.

A Denormalized View – More and more tools for reporting and analytics prefer to have the data organized as

one wide table containing all the relevant columns. Especially many of the newer self‐service data
visualization tools prefer this. They read all that data in the memory of the client machine, turn it
internally into some fast indexed structure, and it’s then presented in a user‐friendly form.

In most cases, such a wide table is a denormalized version of a SuperNova link table, or in fact a
denormalization of an extended SuperNova link table. This link table is joined with all relevant extended
SuperNova tables:

CREATE VIEW DD_LINK AS
SELECT LINK_ID, HUB1_ID, HUB1_STARTDATE, SUPERNOVA_HUB1.ENDDATE AS HUB1_ENDDATE,
HUB2_ID, HUB2_STARTDATE, SUPERNOVA_HUB2.ENDDATE AS HUB2_ENDATE, EVENTDATE,
SUPERNOVA_HUB1.BUSINESS_KEY AS HUB1_BUSINESS_KEY,
SUPERNOVA_HUB1.ATTRIBUTE1 AS HUB1_ATTRIBUTE1,
SUPERNOVA_HUB1.ATTRIBUTE2 AS HUB1_ATTRIBUTE2,
SUPERNOVA_HUB2.BUSINESS_KEY AS HUB2_BUSINESS_KEY,
SUPERNOVA_HUB2.ATTRIBUTE1 AS HUB2_ATTRIBUTE1
FROM SUPERNOVA_LINK_WIDE
ON SUPERNOVA_LINK_WIDE.HUB1_ID = SUPERNOVA_HUB1.HUB_ID AND
SUPERNOVA_LINK_WIDE.HUB1_STARTDATE = SUPERNOVA_HUB1.STARTDATE
ON SUPERNOVA_LINK_WIDE.HUB2_ID = SUPERNOVA_HUB2.HUB_ID AND
SUPERNOVA_LINK_WIDE.HUB2_STARTDATE = SUPERNOVA_HUB2.STARTDATE

Evidently, such a link table can be set up differently. For example, it now contains all the startdates and
enddates. These can be removed to make the table easier to understand for users.

A Star Schema View – The structure of the SuperNova link and hub tables form a star schema pattern,
where the links are facts and the hubs the dimensions. It may be necessary to simplify this structure by
removing columns that are irrelevant for the users, or by showing in the hubs only current versions.

It may also be useful to combine multiple links together to form one dimension. For example, it may be
useful for users to have the links sales per product and number products returned to be combined in one
dimension. Careful that when link tables are joined together that the respective keys match.

A Service View – Most tools use the SQL interface when accessing data. But there are BI applications that
access data through a service interface using SOAP or REST. It’s recommended to determine first which
interfaces they prefer and then determine the structure of a view that can support that. Next, that view is
defined on the existing (SN views. By accessing these, no logic is bypassed and there is a greater chance
that the results obtained via the service interface are in synch with those obtained via the SQL interface.

A Final Remark – It’s not the intention to create a set of views for each report separately. The focus should
be on reuse. Try to implement all the logic in the SuperNova views or the XSN views and as little in the
Data Delivery layer. Use the latter for combining and restructuring purposes.


10 Getting Started

How to get started with SuperNova depends on whether an existing Data Vault already exists or a new
one is being developed (green field situation).

Migrating an Existing Data Vault Environment – If a Data Vault‐based data warehouse environment consisting
of many derived data stores, it’s recommended to select an approach where the existing derived data
stores are gradually replaced by views defined in CIS. The following migration process is proposed:

 Install CIS and connect it to the Data Vault.
 Select a report that must be migrated first (depending on user priorities).
 Select the tables in the derived data stores required by that report.
 Define the SuperNova, Extended SuperNova, and the Data Delivery views to recreate the tables of
the derived data stores.
 If needed, define caches for some of these views.
 Redirect the report to the Data Delivery views and test whether the results are consistent with the
previous solution.
 Extract logic from the reports and move that to the Extended SuperNova views or the Data
Delivery views.
 When all works fine, migrate the next report.

A Green Field Situation – In a green field situation deploy a step‐by‐step, iterative process as well:

 Install CIS and connect it to the database that will contain the Data Vault.
 Design your first set of Data Vault tables and apply the same rules as specified in Section 4.
 Define SuperNova views and Extended SuperNova views for all the Data Vault tables.
 Select a report that must be implemented first (depending on user priorities).
 Define the required Data Delivery views for this report, and implement as much of the logic in the
views and not in the reports. If logic can be reused by other reports, implement it in the Extended
SuperNova views.
 If needed, define caches for some of the views.
 When all works fine, implement the next report.


About the Author Rick F. van der Lans

Rick F. van der Lans is an independent analyst, consultant, author, and lecturer specializing in data
warehousing, business intelligence, database technology, and data virtualization. He works for
R20/Consultancy (www.r20.nl), a consultancy company he founded in 1987.

Rick is chairman of the annual European Enterprise Data and Business Intelligence Conference (organized
annually in London). He writes for the SearchBusinessAnalytics.Techtarget.com, B‐eye‐Network.com8 and
other websites. He introduced the business intelligence architecture called the Data Delivery Platform in
2009 in a number of articles9 all published at BeyeNetwork.com. The Data Delivery Platform is an
architecture based on data virtualization.

He has written several books on SQL. Published in 1987, his popular Introduction to SQL10 was the first
English book on the market devoted entirely to SQL. After more than twenty five years, this book is still
being sold, and has been translated in several languages, including Chinese, German, and Italian. His latest
book11 Data Virtualization for Business Intelligence Systems was published in 2012.

For more information please visit www.r20.nl, or email to rick@r20.nl. You can also get in touch with him
via LinkedIn and via Twitter @Rick_vanderlans.

About Cisco Systems, Inc.

Cisco (NASDAQ: CSCO) is the worldwide leader in IT that helps companies seize the opportunities of
tomorrow by proving that amazing things can happen when you connect the previously
unconnected. Cisco Information Server is agile data virtualization software that makes it easy for
companies to access business data across the network as if it were in a single place.

About Centennium

Centennium specializes in Business Intelligence and Data Warehousing and is a leader in Data Vault based
Business Intelligence solutions. Centennium is the Dutch partner of Genesee Academy and provides
various Data Vault classes and Data Vault Certification CDVDM.

8
See http://www.b‐eye‐network.com/channels/5087/articles/
9
See http://www.b‐eye‐network.com/channels/5087/view/12495
10
R.F. van der Lans, Introduction to SQL; Mastering the Relational Database Language, fourth edition, Addison‐Wesley, 2007.
11
R.F. van der Lans, Data Virtualization for Business Intelligence Systems, Morgan Kaufmann Publishers, 2012.

Whitepaper Cisco Datavault

Uploaded by

Copyright:

Available Formats

Whitepaper Cisco Datavault

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Whitepaper Cisco Datavault

Uploaded by

Copyright:

Available Formats

Data Vault and Data Virtualization:

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

2 Data Vault in a Nutshell

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

Data Virtualization – The challenge for many organizations using Data Vault is to develop an environment

Copyright © 2014 R20/Consultancy, all rights reserved.

3 The SuperNova Data Modeling Technique

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

Layer 4: Extended SuperNova Layer – The Extended SuperNova (XSN) layer consists of a set of views defined

4 Rules For the Data Vault Model

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

5 The Sample Data Vault

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

6 Developing the SuperNova Model

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

For each satellite of the hub the values of the identifier, the startdate and enddate are selected. These

Copyright © 2014 R20/Consultancy, all rights reserved.

3 2013-11-12 00:00:00 2013-11-12 00:00:00 b3

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

LEFT OUTER JOIN SUPERNOVA_HUB1

Copyright © 2014 R20/Consultancy, all rights reserved.

7 Optimizing the SuperNova Model

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

8 Developing the Extended SuperNova Model

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

9 Developing the Data Delivery Model

Copyright © 2014 R20/Consultancy, all rights reserved.

A Denormalized View – More and more tools for reporting and analytics prefer to have the data organized as

Copyright © 2014 R20/Consultancy, all rights reserved.

Copyright © 2014 R20/Consultancy, all rights reserved.

About the Author Rick F. van der Lans

About Cisco Systems, Inc.

Copyright © 2014 R20/Consultancy, all rights reserved.

You might also like