Data Virtualization: Going Beyond Traditional Data Integration to Achieve Business Agility is the first book published on the topic of data virtualization. Along with an overview of data virtualization and its advantages, it presents ten case studies of organizations that have adopted data virtualization to significantly improve business decision making, decrease time-to-solution and reduce costs. This article describes data virtualization adoption at one of the enterprises profiled, Pfizer Inc.

Organization Background
Pfizer Inc. is a biopharmaceutical company that develops, manufactures and markets medicines for both humans and animals. As the world's largest drug manufacturer, Pfizer operates globally with 111,500 employees and a presence in over 100 countries.

Worldwide Pharmaceutical Sciences (PharmSci) is a group of scientists responsible for enabling what drugs Pfizer will bring to market. This group designs, synthesizes and manufactures all drugs that are part of clinical trials and toxicology testing within Pfizer.

For this case study, we interviewed Dr. Michael C. Linhares, Ph.D and Research Fellow. Linhares heads up the Business Information Systems (BIS) team within PharmSci.

BIS is responsible for portfolio and resource management across all of PharmSci's projects. This involves designing, building and supporting systems that deliver data to executive teams and staff to help them make decisions regarding how to allocate available resources - both people and dollars - across the overall portfolio of over 100 projects annually.

The Business Problem
A major challenge for PharmSci is the fact that it has a complex portfolio of projects that is constantly changing.

According to Linhares, "Every week, something new comes up and we need to ensure that the right information is communicated to the right people. The people making decisions about resource allocation need easy and simple methods for obtaining that information. One aspect of this is that some people learn the information first and they need to communicate it to others who are responsible for making decisions based on the information. This creates an information-sharing challenge."

Linhares estimates that there are 80 to 100 information producers within PharmSci and over 1,000 information consumers, including the executives who seek a full picture of the project portfolio - financial data, project data, people data and data about the pharmaceutical compounds themselves.

The Technical Problem
The data required is created in and managed by different applications, each developed by a different team, stored in multiple sources managed by different technologies, and the applications don't talk to each other.

This makes it very difficult to access summary information across all projects. Examples would be identifying how much money is being spent on all projects in the project management system, what the next milestones are and when each will be met, and who is working on each project. "We needed a solution that would allow us to pull all this information together in an agile way."

When Linhares joined PharmSci, there was very little in the way of effective information integration. Most integration was done manually by exporting data from various systems into Excel spreadsheets and then either combining spreadsheets or taking the spreadsheet data and moving it into Access or SQL Server databases. With no real security controls, this approach also lacked scalability and opportunities for reuse, generated multiple copies of the spreadsheets (with various changes), and it often took weeks to build a spreadsheet with only a 50% chance that it would include all of the data required.

Solution Requirements
To be successful, the solution to these data integration and reporting problems had to provide the following:

• A single, integrated view of all data sources with a common set of naming conventions
• A flexible middle layer that would be independent of both the data sources on the back end and the reporting tools on the front end to facilitate easy change management
• Shared metadata and business rule functionality so there would be a single point for managing and monitoring the solution
• A development platform that supported fast, iterative development and, therefore, continuous process improvement

Three Options Considered
BIS considered three solution architectures to meet their business and technical challenges.

1. Traditional Information Factory: The first option was a traditional approach of an integrated, scalable information factory. Pfizer had already implemented information factories in the division using a combination of Informatica ETL tools, Oracle databases and custom-built reporting applications. However, according to Linhares, an information factory "seemed like overkill. We didn't have high volumes of data, nor did we need the inherent complexity of using ETL tools to transform and move data while making sure we included all the detailed data we might possibly ever need over time." Furthermore, because of the way the information factories were managed within Pfizer, change management entailed significant overhead. However, the architectural concepts of an information factory were not going to be ignored in the final solution.
2. Single Vendor Stack: A second possible approach was to implement the solution in a single integrated technology (SQL Server with integration services). Major disadvantages were the lack of access to multiple data source types, the need to move data multiple times and the lack of an integrated metadata repository for understanding and organizing the data model.
3. Data Virtualization: The third option was to create a federated data virtualization layer that integrated and accessed the underlying data sources through virtual views of the data. By leaving the source data in place, this approach would eliminate the issues inherent in copying and moving all the data (which Linhares described as unnecessary, "non-value added" activities). With the right technology and mix of products, data virtualization would enable PharmSci to migrate from inefficient, off-line spreadmarts to online access to integrated information that could be rapidly tailored and reused to dramatically increase its value to the organization.

The Data Virtualization Solution - Architecture
Pfizer's solution is the PharmSci Portfolio Database (PSPD), a federated data delivery framework implemented with the Composite Data Virtualization Platform.

Data virtualization enables the integration of all PharmSci data sources into a single reporting schema of information that can be accessed by all front-end tools and users. The solution architecture includes the following components:

Trusted Data Sources: There are many sources of data for PSPD; they are geographically dispersed, store data in a variety of formats across a multivendor, heterogeneous data environment. Here are some examples:

• Enterprise Project Management (EPM) is a SQL Server database of WRD's drug portfolio project plans. It includes detailed project schedules and milestones.
• The Global Information Factory (GIF) is an Oracle-based data warehouse of monthly finance data.
• OneSource, a database of corporate-level drug portfolio information is itself a unified set of Composite views across several different sources built by another group within Pfizer.
• Flat files are provided by the Finance Department on actual resource use.
• SharePoint lists are small SharePoint databases accessed using a web service.
• There are other data sources as well, including custom-built systems. As Linhares pointed out, "It doesn't matter what data sources we have. With a virtual approach, we are not limited by the types of data we need to access."

Data Virtualization Layer: The Composite Data Virtualization Platform forms the data virtualization layer that enables the solution to be independent of the data sources and front-end tools. It provides abstracted access to all of the data sources and delivers the data through virtual views. These views effectively present the PharmSci Portfolio Database as subject-specific data marts. The Composite metadata repository manages data lineage and business rules.

Consuming Applications: The flexibility of the platform is demonstrated by the varied reporting applications that use the information in PSPD. Examples include:

• SAP Business Objects for ad hoc queries, standard reports and dashboards.
• TIBCO Spotfire for analytics and access to data through standard presentation reports.
• Web services for parameterized queries.
• Data services to provide data for downstream applications.
• QuickViews (web pages built using DevExpress, a .NET toolkit) for access to live data.

SharePoint Portal: Branded as "InfoSource," this team collaboration web portal is the front-end interface that provides integrated access to PSPD data for all PharmSci customers through the consuming applications described above.

The Data Virtualization Solution - Best Practices
Linhares and team applied a number of data virtualization best practices when implementing the architecture described above.

Two Layers of Abstraction: Linhares stressed the importance of building two clear levels of abstraction into the data virtualization architecture. The first level abstracts Sources (the information abstraction layer), the second consumers (the reporting abstraction layer).

"We built a representation of the data in Composite. If a source is ever changed by the owner, which often happens, we can update the representation in the information abstraction layer quickly. This allows control of all downstream data in one location."

The second level of abstraction is the one between the reporting schema and the front-end reporting tools. A consolidated and integrated set of information is exposed as a single schema. This allows BIS to be system agnostic and support the use of whatever tool is best for the customer. All of the reporting tools use the same reporting abstraction layer; they always get the same answer to the same question because there is only a single source of data.

Consolidated Business Rules: Another key piece of the solution is the ability to include the business rules about how PharmSci manages its data within these abstraction layers. The business rules are embedded in the view definitions and are applied consistently at the same point.

Rapid Application Development Process: Prior to data virtualization, data integration was the slowest step for BIS in fulfilling a customer request for information. Now it's typically the fastest. "For example, a request that came in Friday morning and was completed by that afternoon. The customer's response was an amazed, ‘What do you mean you already have it done?'"

BIS uses a simple development process. The first step is what Linhares calls "triage" - looking at what the customer wants, estimating how long it will take and communicating that to the customer.

BIS does not spend a lot of time documenting the requirements of the solution. Instead, the group first creates a prototype on paper in the form of a simple data flow, then creates the necessary virtual views, gives the customer web access to the views and asks: "Is this what you wanted?"

The customer can then play with the result and respond with any changes or additions needed. BIS arrives at the final solution working with the customer in an iterative process.

Summary of Benefits
Linhares described several major benefits of the data virtualization solution.

The ability to provide integrated data in context: Data virtualization has enabled BIS to replace isolated silos of data with a data delivery platform that integrates different types and sources of data into a comprehensive package of value-added information. Instead of only the team leader and a core group of eight to ten people knowing about a project, the entire organization has access to relevant project information.

The independence of the data virtualization layer: "This is one of the huge benefits of data virtualization. It allows me to manage and monitor everything in one place and it makes change management easy for BIS and transparent to users."

Fast, iterative development environment: The data delivery infrastructure already exists in the data virtualization layer (defined data sources, standard naming conventions, access methods, etc.) so when a request for information comes in, BIS can quickly put it together for the customer.

Elimination of manual effort throughout PharmSci: According to Linhares, people initially resisted going away from their spreadsheets. But once there was a single source for the data and it was all available through InfoSource, there was a dramatic reduction in the need to have meetings to reconcile spreadsheet data among teams.

•   •   •

Editor's Note: Robert Eve is the co-author, along with Judith R. Davis, of Data Virtualization: Going Beyond Traditional Data Integration to Achieve Business Agility, the first book published on the topic of data virtualization. The complete Pfizer case study, along with nine others enterprise are available in the book.

• How Data Virtualization Improves Business Agility – Part 3
• Data Virtualization at Northern Trust: A Case Study
• How Data Virtualization Delivers Business Agility – Part 1
• How Data Virtualization Improves Business Agility – Part 2