An Introduction to SQL Server 2005 Integration Services
This paper discusses the challenges that face businesses that rely on data integration technologies to provide meaningful, reliable information to maintain a competitive advantage in today’s business world. It discusses how SQL Server 2005 Integration Services (SSIS) can help Information Technology departments meet data integration requirements in their companies. Real-world scenarios are included.
On This Page
The ability to transform corporate data into meaningful and actionable information is the single most important source of competitive advantage in today’s business world. Harnessing the data explosion to better understand the past and get direction for the future has turned out to be one of the most challenging ventures for enterprise Information Technology departments in global organizations. There are three broad categories of issues associated with data integration:
In this paper, we will explore these challenges in detail and discuss how to address them with Microsoft® SQL Server™ 2005 Integration Services (SSIS). First, let’s view them in the context of a real-world scenario.
A Real-World Scenario
A major global transportation company uses its data warehouse to both analyze the performance of its operations and to predict variances in its scheduled deliveries.
The major sources of data in this company include order data from its DB2-based order entry system, customer data from its SQL Server-based customer relationship management (CRM) system, and vendor data from its Oracle-based ERP system. In addition to data from these major systems, data from spreadsheets tracking “extraordinary” events, which have been entered by hand by shipping supervisors, is incorporated into the data warehouse. Currently, external data such as weather information, traffic status, and vendor details (for subcontracted deliveries) are incorporated on a delayed basis from text files from various sources.
Not only are the sources for these data diverse, but the consumers are also diverse both in their requirements and their geographic locations. This diversity has led to a proliferation of local systems. One of the major efforts for the Information Technology department is to establish a “single version of the truth,” at least for its customer data.
Data Integration Requirements
In view of this diversity of data, business needs, and user requirements, the Information Technology department has specified the following set of data integration requirements:
They must provide reliable and consistent historical and current data integrated from a variety of internal and external sources.
To reduce lags in data acquisition, data from providers and vendors must be available via Web services or some other direct mechanism such as FTP.
They need to cleanse and remove duplicate data and otherwise enforce data quality.
Increasing global regulatory demands require that the company maintain clear audit trails. It is not enough to maintain reliable data; the data needs to be tracked and certified.
Challenges of Data Integration
At one level, the problem of data integration in our real-world scenario is extraordinarily simple. Get data from multiple sources, cleanse and transform the data, and load the data into appropriate data stores for analysis and reporting. Unfortunately, in a typical data warehouse or business intelligence project, enterprises spend 60–80% of the available resources in the data integration stage. Why is it so difficult?
Technology challenges start with source systems. We are moving from collecting data on transactions (where customers commit to getting, buying, or otherwise acquiring something) to collecting data on pre-transactions (where customer intentions are tracked via mechanisms such as Web clicks or RFID). Data is now not only acquired via traditional sources and formats, such as databases and text files, but is increasingly available in a variety of different formats (ranging from proprietary files to Microsoft Office documents to XML-based files) and from Internet-based sources such as Web services and RSS (Really Simple Syndication) streams. The most pertinent challenges are:
Multiple sources with different formats.
Structured, semi-structured, and unstructured data.
Data feeds from source systems arriving at different times.
Huge data volumes.
In an ideal world, even if we somehow manage to get all the data we need in one place, new challenges start to surface, including:
Making sense of different data formats.
Transforming the data into a format that is meaningful to business analysts.
Suppose that we can magically get all the data we need and that we can cleanse, transform, and map the data into a useful format. There is still another shift away from traditional data movement and integration. That is the shift from fixed long batch-oriented processes to fluid and shorter on-demand processes. Batch-oriented processes are usually performed during “downtimes” when users do not place heavy demands on the system. This usually is at night during a predefined batch window of 6-8 hours, when no one is supposed to be in the office. With the increasing globalization of businesses of every size and type, this is no longer true. There is very little (if any) downtime and someone is always in the office somewhere in the world. The sun really doesn’t set on the global business.
As a result we have:
Increasing pressure to load the data as quickly as possible.
The need to load multiple destinations at the same time.
Not only do we need to do all these, but we need to do them as fast as possible. In extreme cases, such as online businesses, data needs to be integrated on a continuous basis. There are no real batch windows and latencies can not exceed minutes. In many of these scenarios, the decision making process is automated with continuously running software.
Scalability and performance become more and more important as we face business needs that can’t tolerate any downtime.
Without the right technology, systems require staging at almost every step of the warehousing and integration process. As different (especially nonstandard) data sources need to be included in the ETL (Extract, Transform, and Load) process and as more complex operations (such as data and text mining) need to be performed on the data, the need to stage the data increases. As illustrated in Figure 1, with increased staging the time taken to “close the loop,” (i.e., to analyze, and take action on new data) increases as well. These traditional ELT architectures (as opposed to value-added ETL processes that occur prior to loading) impose severe restrictions on the ability of systems to respond to emerging business needs.
Finally, the question of how data integration ties into the overall integration architecture of the organization is becoming more important when both the real-time transactional technology of application integration and the batch-oriented high-volume world of data integration technology are needed to solve the business problems of the enterprise.
There are two broad issues with data integration in a large organization; these are the “power” problem, and the “comfort zone” problem.
Power Challenge: Data is power and it is usually very hard to make people think of data in terms of a real valuable shared asset of the company. For enterprise data integration to be successful, all the owners of multiple data sources have to whole-heartedly buy into the purpose and direction of the project. Lack of cooperation from the relevant parties is one the major reasons for the failure of data integration projects. Executive sponsorship, consensus building, and a strong data integration team with multiple stakeholders are a few of the critical success factors that can help resolve the issues.
Comfort Zone Challenge: Problems of data integration, when analyzed in the context of an isolated need, can be solved in multiple ways. About 60% of data integration is solved by hand-coding. The technology used to solve similar problems can range from replication, ETL, SQL, to EAI. People gravitate towards the technology they are familiar with. Although these approaches have overlapping capabilities and can perhaps do the job in isolated cases, these technologies are optimized to solve different sets of problems. When trying to solve the problem of enterprise data integration, the lack of a sound architecture with appropriate technology choices can turn out to be a recipe for failure.
The organizational and technology related issues previously outlined conspire together to make data integration the most expensive part of any data warehouse/business intelligence project. The major factors that add to the cost of data integration are:
Getting the data out in the format that is necessary for data integration ends up being a slow and torturous process fraught with organizational power games.
Cleansing the data and mapping the data from multiple sources into one coherent and meaningful format is extraordinarily difficult
More often than not, standard data integration tools don’t offer enough functionality or extensibility to satisfy the data transformation requirements for the project. This can result in the expenditure of large sums of money in consulting costs to develop special ETL code to get the job done.
Different parts of the organization focus on the data integration problem in silos.
When there is a need to put them all together, additional costs are incurred to integrate these efforts into an enterprise-wide data integration architecture.
As the data warehousing and business intelligence needs of the organization evolve, faulty data integration architecture becomes more and more difficult to maintain and the total cost of ownership skyrockets.
SQL Server 2005 Integration Services
The traditional ETL-centric data integration from standard data sources continues to be at the heart of most data warehouses. However, demands to include more diverse data sources, regulatory requirements, and global and online operations are quickly transforming the traditional requirements for data integration. In this fast growing and changing landscape, the need to extract value from data and the need to be able to rely on it is more important than ever before. Effective data integration has become the basis of effective decision making. SQL Server Integration Services provides a flexible, fast, and scalable architecture that enables effective data integration in current business environments.
In this paper we will examine how SQL Server Integration Services (SSIS) is an effective toolset for both the traditional demands of ETL operations, as well as for the evolving needs of general purpose data integration. We will also discuss how SSIS is fundamentally different from the tools and solutions provided by major ETL vendors so it is ideally suited to address the changing demands of global business from the largest enterprise to the smallest business.
Task flow and data flow engine
SSIS consists of both an operations-oriented task-flow engine as well as a scalable and fast data-flow engine. The data flow exists in the context of an overall task flow. It is the task-flow engine that provides the runtime resource and operational support for the data-flow engine. This combination of task flow and data flow enables SSIS to be effective in traditional ETL or data warehouse (DW) scenarios as well as in many other extended scenarios such as data center operations. In this paper we will mainly focus on the data-flow related scenarios. The use of SSIS for data center oriented workflow is a separate topic by itself.
At the core of SSIS is the data transformation pipeline. This pipeline has a buffer-oriented architecture that is extremely fast at manipulating rowsets of data once they have been loaded into memory. The approach is to perform all data transformation steps of the ETL process in a single operation without staging data, although specific transformation or operational requirements, or indeed hardware may be a hindrance. Nevertheless, for maximum performance, the architecture avoids staging. Even copying the data in memory is avoided as far as possible. This is in contrast to traditional ETL tools, which often require staging at almost every step of the warehousing and integration process. The ability to manipulate data without staging extends beyond traditional relational and flat file data and beyond traditional ETL transformation capabilities. With SSIS, all types of data (structured, unstructured, XML, etc.) are converted to a tabular (columns and rows) structure before being loaded into its buffers. Any data operation that can be applied to tabular data can be applied to the data at any step in the data-flow pipeline. This means that a single data-flow pipeline can integrate diverse sources of data and perform arbitrarily complex operations on these data without having to stage the data.
It should also be noted though, that if staging is required for business or operational reasons, SSIS has good support for these implementations as well.
This architecture allows SSIS to be used in a variety of data integration scenarios, ranging from traditional DW-oriented ETL to nontraditional information integration technologies.
SSIS for Traditional DW Loading
At its core, SSIS is a comprehensive, fully functional ETL tool. Its functionality, scale, and performance compare very favorably with high-end competitors in the market at a fraction of their cost. The data integration pipeline architecture allows it to consume data from multiple simultaneous sources, perform multiple complex transformations, and then land the data to multiple simultaneous destinations. This architecture allows SSIS to be used not only for large datasets, but also for complex data flows. As the data flows from source(s) to destination(s), the stream of data can be split, merged, combined with other data streams, and otherwise manipulated. Figure 2 shows an example of such a flow:
SSIS can consume data from (and land data into) a variety of sources including OLE DB, managed (ADO.NET), ODBC, flat file, Excel, and XML using a specialized set of components called adapters. SSIS can even consume data from custom data adapters (developed in-house or by third parties). This allows the wrapping of legacy data loading logic into a data source that can be seamlessly consumed in the SSIS data flow. SSIS includes a set of powerful data transformation components that allow data manipulations that are essential for building data warehouses. These transformation components include:
Aggregate Performs multiple aggregates in a single pass.
Sort Sorts data in the flow.
Lookup Performs flexible cached lookup operations to reference datasets.
Pivot and UnPivot Two separate transformations do exactly as their names suggest.
Merge, Merge Join, and UnionAll Can perform join and union operations.
Derived Column Performs column-level manipulations such as string, numeric, date/time, etc. operations, and code page translations. This one component actually wraps what other vendors might break up into many different transformations.
Data Conversion Converts data between various types (numeric, string, etc.).
Audit Adds columns with lineage metadata and other operational audit data.
In addition to these core data warehousing transformations, SSIS includes support for advanced data warehousing needs such as Slowly Changing Dimensions (SCD). The SCD Wizard in SSIS guides users through specifying their requirements for managing slowly changing dimensions and, based upon their input, generates a complete data flow with multiple transformations to implement the slowly changing dimension load. Support for standard Type 1 and 2 SCD along with 2 new SCD types (Fixed Attributes and Inferred Members) is provided. Figure 3 shows a page from the SCD Wizard.
Figure 4 shows the data flow generated by this Wizard.
SSIS also can be used to load Analysis Services multidimensional OLAP (MOLAP) caches directly from the data-flow pipeline. This means that SSIS can not only be used to create relational data warehouses, but also to load multidimensional cubes for analytical applications.
SSIS and Data Quality
One of the key features of SSIS is its ability to not only integrate data, but also to integrate different technologies to manipulate the data. This has allowed SSIS to include cutting edge “fuzzy logic” based data cleansing components. These components were developed by the Microsoft Research labs and represent the latest research in this area. The approach taken is a domain independent one and doesn’t depend upon any specific domain data, such as address/zip reference data. This allows these transformations to be used for cleansing most types of data, not just address data.
SSIS is deeply integrated with the data mining functionality in Analysis Services. Data mining abstracts out the patterns in a dataset and encapsulates them in a mining model. This mining model amongst other things then can be used to make predictions on what data belongs to a dataset and what data may be anomalous, allowing data mining to be used as a tool for implementing data quality. Support for complex data routing in SSIS allows anomalous data to not only be identified, but also be automatically corrected and replaced with better values. This enables “closed loop” cleansing scenarios. Figure 5 shows an example of such a closed loop cleansing data flow.
In addition to its built-in data quality features, SSIS can be extended to work closely with third-party data-cleansing solutions.
Application of SSIS Beyond Traditional ETL
The ability of the data-flow pipeline to manipulate almost any kind of data, the deep integration with Analysis Services, the support for extending it with a large variety of data manipulation technologies, and the inclusion of a rich work-flow engine allow SSIS to be used in many scenarios that are not traditionally thought of as ETL
Service Oriented Architecture
SSIS includes support for sourcing XML data in the data-flow pipeline, including data both from files on disk as well as URLs over HTTP. XML data is “shredded” into tabular data, which then can be easily manipulated in the data flow. This support for XML can work with the support for Web services. SSIS can interact with Web services in the control flow to capture XML data.
XML can also be captured from files, from Microsoft Message Queuing (MSMQ), and over the Web via HTTP. SSIS enables the manipulation of the XML with XSLT, XPATH, diff/merge, etc. and can also stream the XML into the data flow
This support enables SSIS to participate in flexible Service Oriented Architectures (SOA).
Data and text mining
SSIS not only has deep integration with the data mining features from Analysis Services, but it also has text mining components. Text mining (also referred to as text classification) involves identifying the relationship between business categories and the text data (words and phrases). This allows the discovery of key terms in text data and based upon this to automatically identify text that is “interesting.” This in turn can drive “closed-loop” actions to achieve business goals such as increasing customer satisfaction and enhancing the quality of the products and services.
On-demand data source
One of the most unique features in SSIS is the DataReader destination, which lands data into an ADO.NET DataReader. When this component is included in a data-flow pipeline, the package containing the DataReader destination can be used as a data source, exposed as an ADO.NET DataReader itself. This allows SSIS to be used not only as a traditional ETL to load data warehouses, but also as a data source that can deliver integrated, reconciled, and cleansed data from multiple sources on-demand. For example, this might be used to allow Reporting Services to consume data from multiple diverse data sources using a SSIS package as its data source.
A possible scenario that integrates all of these, consists of identifying and delivering interesting articles from RSS feeds as part of a regular report. Figure 6 shows a SSIS package that sources data from RSS feeds over the Internet, integrates with data from a Web service, performs text mining to find interesting articles from the RSS feeds, and then lands the interesting articles into a DataReader destination to be finally consumed by a Reporting Services report.
Figure 7 shows the use of the SSIS package as a data source in the Report Wizard.
From an ETL tool perspective, this scenario is very unusual because there really isn’t any data extraction or transformation or loading.
SSIS, the Integration Platform
SSIS goes beyond being an ETL tool not only in terms of enabling nontraditional scenarios, but also in being a true platform for data integration. SSIS is part of the SQL Server Business Intelligence (BI) platform which enables the development of end-to-end BI applications.
Integrated development platform
SQL Server Integration Services, Analysis Services, and Reporting Services all use a common Visual Studio® based development environment called the SQL Server Business Intelligence (BI) Development Studio. BI Development Studio provides an integrated development environment (IDE) for BI application development. This shared infrastructure enables metadata-level integration between various development projects (integration, analysis, and reporting). An example of such shared construct is the Data Source View (DSV), which is an offline schema/view definition of data sources, and is used by all three BI project types.
This IDE provides facilities such as integration with version control software (e.g., VSS) along with support for team-based features such as “check-in/check-out” and as such it fulfills the need for an enterprise-class team-oriented development environment for business intelligence applications. Figure 8 shows a BI Development Studio solution that consists of Integration, Analysis, and Reporting projects.
Not only does this provide a single place to develop BI applications, but it also can be used to develop other Visual Studio projects (using Visual C#®, Visual Basic® .NET etc.) and so can provide developers with a true end-to-end development experience.
Besides an integrated BI development environment, BI Development Studio has features for true run-time debugging of SSIS packages. These include the ability to set breakpoints and support for standard development constructs such as watching variables. A truly unique feature is the Data Viewer, which provides the ability to view rows of data as they are processed in the data-flow pipeline. This visualization of data can be in the form of a regular text grid or a graphical presentation such as a scatter plot or bar graph. In fact, it is possible to have multiple connected viewers that can display the data simultaneously in multiple formats. Figure 9 shows an example of geographic data visualized using a scatter plot and a text grid.
In addition to providing a professional development environment, SSIS exposes all its functionality via a set of rich APIs. These APIs are both managed (.NET Framework) and native (Win32) and allow developers to extend the functionality of SSIS by developing custom components in any language supported by the .NET Framework (such as Visual C#, Visual Basic .NET, etc.) and C++. These custom components can be work-flow tasks and data-flow transformations (including source and destination adapters). This allows legacy data and functionality to be easily included in SSIS integration processes, allowing the past investments in legacy technologies to be effectively leveraged. It also allows easy inclusion of third-party components.
The extensibility previously mentioned is not only limited to re-usable custom components but also includes script-based extensibility. SSIS has script components both for task flow as well as for data flow. These allow users to write scripts in Visual Basic. NET to add ad hoc functionality (including data sources and destinations) and to re-use any preexisting functionality packaged as .NET Framework assemblies.
Figure 10 shows an example of a script that manipulates rows of data inside a data flow.
This extensibility model makes SSIS not only a data integration tool, but also an Integration Bus into which technologies like data mining, text mining, and UDM can easily be plugged in to enable complex integration scenarios involving pretty much arbitrary data manipulation and structures.
Making Data Integration Approachable
The flexible and extensible architecture of SSIS allows it to address most of the technology challenges to data integration outlined earlier in this paper. As shown in Figure 11, SSIS eliminates (or at least minimizes) unnecessary staging. Because it performs complex data manipulation in a single pipeline operation, it is now possible to react to changes and patterns in the data fairly quickly, in a time frame that is actually meaningful for closing the loop and taking action. This is in contrast to traditional architectures that rely on data staging and that become impractical for closing the loop and taking meaningful action on data.
The extensible nature of SSIS makes it possible for organizations to leverage their existing investments in custom code for data integration by wrapping it as re-usable extensions to SSIS and by doing so to take full advantage of features such as logging, debugging, BI integration, etc. This greatly helps to overcome some of the organizational challenges outlined earlier in this paper.
The inclusion of SSIS in the SQL Server product makes the cost acquisition extremely reasonable as compared to other high-end data integration tools. Not only is the initial cost acquisition lowered, but via tight integration with Visual Studio and the rest of SQL Server BI tools, the cost of application development and maintenance is also significantly lowered in comparison to other similar tools. The extremely reasonable total cost of ownership (TCO) of SSIS (and the rest of SQL Server) makes enterprise-class data integration approachable to all segments of the market, taking it out of the exclusive domain of the largest (and richest) companies. At the same time, the architecture of SSIS is tuned to take advantage of modern hardware and to deliver performance and scale at the highest end of customer requirements. SSIS enables rich, scalable data integration to all customers, from the highest end enterprise to the small and medium business. In conjunction with the rest of the features in SQL Server, the Microsoft customer support infrastructure (ranging from broad, long beta testing, to rich online communities to premiere support contracts) and the consistency and integration with the rest of Microsoft product offerings, SSIS is truly a unique toolset that opens up new frontiers in data integration.