The Extract, Transformation, and Load (ETL) system is the most time-consuming and expensive part of building a data warehouse and delivering business intelligence to your user community. A decade ago the majority of ETL systems were hand crafted, but the market for ETL software has steadily grown and the majority of practitioners now use ETL tools in place of hand-coded systems.
Does it make sense to hand-code an ETL system in 2008, or is an ETL tool a better choice? Kimball Group now generally recommends using an ETL tool, but a custom-built approach can still makes sense. This article summarizes the advantages and disadvantages of ETL tools and offers advice on making the choice that's right for you.
ADVANTAGES OF ETL TOOLS
Visual flow and self-documentation.The single greatest advantage of an ETL tool is that it provides a visual flow of the system's logic. Each ETL tool presents these flows differently, but even the least-appealing of these user interfaces compare favorably to custom systems consisting of stored procedures, SQL and operating system scripts, and a handful of other technologies. Ironically, some ETL tools have no practical way to print the otherwise-attractive self documentation.
Learn about a predictive analytics solution that provides forward-looking insights, and offers best practices to ensure you maximize the value of your company's customer relationships
Best Practices for Collaboration in the Enterprise
Structured system design. ETL tools are designed for the specific problem of populating a data warehouse. Although they are only tools, they do provide a metadata-driven structure to the development team. This is particularly valuable for teams building their first ETL system.
Operational resilience. Many of the home-grown ETL systems I've evaluated are fragile: they have too many operational problems. ETL tools provide functionality and practices for operating and monitoring the ETL system in production. You can certainly design and build a well instrumented hand-coded ETL application, and ETL tool operational features have yet to mature. Nonetheless, it's easier for a data warehouse / business intelligence team to build on the management features of an ETL tool to build a resilient ETL system.
Data-lineage and data-dependency functionality. We would like to be able to right-click on a number in a report and see exactly how it was calculated, where the data was stored in the data warehouse, how it was transformed, when the data was most recently refreshed, and what source system or systems underlay the numbers. Dependency is the flip side of lineage: we'd like to look at a table or column in the source system and know which ETL modules, data warehouse tables, OLAP cubes, and user reports might be affected by a structural change. In the absence of ETL standards that hand-coded systems could conform to, we must rely on ETL tool vendors to supply this functionality — though, unfortunately, few have done so to date.
Advanced data cleansing functionality. Most ETL systems are structurally complex, with many sources and targets. At the same time, requirements for transformation are often fairly simple, consisting primarily of lookups and substitutions. If you have a complex transformation requirement, for example if you need to de-duplicate your customer list, you should use a specialized tool. Most ETL tools either offer advanced cleansing and de-duplication modules (usually for a substantial additional price) or they integrate smoothly with other specialized tools. At the very least, ETL tools provide a richer set of cleansing functions than are available in SQL.
Performance. You might be surprised that performance is listed last under the advantages of the ETL tools. It's possible to build a high-performance ETL system whether you use a tool or not. It's also possible to build an absolute dog of an ETL system whether you use a tool or not. I've never been able to test whether an excellent hand-coded ETL system outperforms an excellent tool-based ETL system; I believe the answer is that it's situational. But the structure imposed by an ETL tool makes it easier for an inexperienced ETL developer to build a quality system.
Software licensing cost.The greatest disadvantage of ETL tools in comparison to hand-crafted systems is the licensing cost for the ETL tool software. Costs vary widely in the ETL space, from several thousand dollars to hundreds of thousands of dollars.
Uncertainty. We've spoken with many ETL teams that are uncertain – and sometimes misinformed – about what an ETL tool will do for them. Some teams under-value ETL tools, believing they are simply a visual way to connect SQL scripts together. Other teams unrealistically over-value ETL tools, imagining that building the ETL system with such a tool will be more like installing and configuring software than developing an application.
Reduced flexibility. A tool-based approach limits you to the tool vendor's abilities and scripting languages. Build a Solid Foundation
Learn about a predictive analytics solution that provides forward-looking insights, and offers best practices to ensure you maximize the value of your company's customer relationships
Best Practices for Collaboration in the Enterprise
There are some over-arching themes in successful ETL system deployments regardless of which tools and technologies are used. Most important — and most frequently neglected — is the practice of designing the ETL system before development begins. Too often we see systems that just evolved without any initial planning. These systems are inefficient and slow, they break down all the time, and they're unmanageable. The data warehouse team has no idea how to pinpoint the bottlenecks and problem areas of the system. A solid system design should incorporate the concepts described in detail in Kimball University: The Subsystems of ETL Revisited, by Bob Becker.
Good ETL system architects will design standard solutions to common problems such as surrogate key assignment. Excellent ETL systems will implement these standard solutions most of the time but offer enough flexibility to deviate from those standards where necessary. There are usually half a dozen ways to solve any ETL problem, and each one may be the best solution in a specific set of circumstances. Depending on your personality and fondness for solving puzzles, this can be either a blessing or a curse.
One of the rules you should try to follow is to write data as seldom as possible during the ETL process. Writing data, especially to the relational database, is one of the most expensive tasks that the ETL system performs. ETL tools contain functionality to operate on data in memory and guide the developer along a path to minimize database writes until the data is clean and ready to go into the data warehouse table. However, the relational engine is excellent at some tasks, particularly joining related data. There are times when it is more efficient to write data to a table, even index it, and let the relational engine perform a join than it is to use the ETL tool's lookup or merge operators. We usually want to use those operators, but don't overlook the powerful relational database when trying to solve a thorny performance problem.
Whether your ETL system is hand-coded or tool-based, it's your job to design the system for manageability, auditability, and restartability. Your ETL system should tag all rows in the data warehouse with some kind of batch identifier or audit key that describes exactly which process loaded the data. Your ETL system should log information about its operations, so your team can always know exactly where the process is now and how long each step is expected to take. You should build and test procedures for backing out a load, and, ideally, the system should roll back transactions in the event of a midstream failure. The best systems monitor data health during extraction and transformation, and they either improve the data or issue alerts if data quality is substandard. ETL tools can help you with the implementation of these features, but the design is up to you and your team.
Should you use an ETL tool? Yes. Do you have to use an ETL tool? No. For teams building their first or second ETL system, the main advantage of visual tools are self-documentation and a structured development path. For neophytes, these advantages are worth the cost of the tool. If you're a seasoned expert — perhaps a consultant who has built dozens of ETL systems by hand — it's tempting to stick to what has worked well in the past. With this level of expertise, you can probably build a system that performs as well, operates as smoothly, and perhaps costs less to develop than a tool-based ETL system. But many seasoned experts are consultants, so you should think objectively about how maintainable and extensible a hand-crafted ETL system might be once the consultant has moved on.
Don't expect to reap a positive return on investment in an ETL tool during the development of your first system. The advantages will come as that first phase moves into operation, as it's modified over time, and as your data warehouse grows with the addition of new business process models and associated ETL systems.
SOURCE:http://www.informationweek.com/news/software/bi/showArticle.jhtml?articleID=207002081&pgno=3&queryText=&isPrev=
Does it make sense to hand-code an ETL system in 2008, or is an ETL tool a better choice? Kimball Group now generally recommends using an ETL tool, but a custom-built approach can still makes sense. This article summarizes the advantages and disadvantages of ETL tools and offers advice on making the choice that's right for you.
ADVANTAGES OF ETL TOOLS
Visual flow and self-documentation.The single greatest advantage of an ETL tool is that it provides a visual flow of the system's logic. Each ETL tool presents these flows differently, but even the least-appealing of these user interfaces compare favorably to custom systems consisting of stored procedures, SQL and operating system scripts, and a handful of other technologies. Ironically, some ETL tools have no practical way to print the otherwise-attractive self documentation.
Learn about a predictive analytics solution that provides forward-looking insights, and offers best practices to ensure you maximize the value of your company's customer relationships
Best Practices for Collaboration in the Enterprise
Structured system design. ETL tools are designed for the specific problem of populating a data warehouse. Although they are only tools, they do provide a metadata-driven structure to the development team. This is particularly valuable for teams building their first ETL system.
Operational resilience. Many of the home-grown ETL systems I've evaluated are fragile: they have too many operational problems. ETL tools provide functionality and practices for operating and monitoring the ETL system in production. You can certainly design and build a well instrumented hand-coded ETL application, and ETL tool operational features have yet to mature. Nonetheless, it's easier for a data warehouse / business intelligence team to build on the management features of an ETL tool to build a resilient ETL system.
Data-lineage and data-dependency functionality. We would like to be able to right-click on a number in a report and see exactly how it was calculated, where the data was stored in the data warehouse, how it was transformed, when the data was most recently refreshed, and what source system or systems underlay the numbers. Dependency is the flip side of lineage: we'd like to look at a table or column in the source system and know which ETL modules, data warehouse tables, OLAP cubes, and user reports might be affected by a structural change. In the absence of ETL standards that hand-coded systems could conform to, we must rely on ETL tool vendors to supply this functionality — though, unfortunately, few have done so to date.
Advanced data cleansing functionality. Most ETL systems are structurally complex, with many sources and targets. At the same time, requirements for transformation are often fairly simple, consisting primarily of lookups and substitutions. If you have a complex transformation requirement, for example if you need to de-duplicate your customer list, you should use a specialized tool. Most ETL tools either offer advanced cleansing and de-duplication modules (usually for a substantial additional price) or they integrate smoothly with other specialized tools. At the very least, ETL tools provide a richer set of cleansing functions than are available in SQL.
Performance. You might be surprised that performance is listed last under the advantages of the ETL tools. It's possible to build a high-performance ETL system whether you use a tool or not. It's also possible to build an absolute dog of an ETL system whether you use a tool or not. I've never been able to test whether an excellent hand-coded ETL system outperforms an excellent tool-based ETL system; I believe the answer is that it's situational. But the structure imposed by an ETL tool makes it easier for an inexperienced ETL developer to build a quality system.
Software licensing cost.The greatest disadvantage of ETL tools in comparison to hand-crafted systems is the licensing cost for the ETL tool software. Costs vary widely in the ETL space, from several thousand dollars to hundreds of thousands of dollars.
Uncertainty. We've spoken with many ETL teams that are uncertain – and sometimes misinformed – about what an ETL tool will do for them. Some teams under-value ETL tools, believing they are simply a visual way to connect SQL scripts together. Other teams unrealistically over-value ETL tools, imagining that building the ETL system with such a tool will be more like installing and configuring software than developing an application.
Reduced flexibility. A tool-based approach limits you to the tool vendor's abilities and scripting languages. Build a Solid Foundation
Learn about a predictive analytics solution that provides forward-looking insights, and offers best practices to ensure you maximize the value of your company's customer relationships
Best Practices for Collaboration in the Enterprise
There are some over-arching themes in successful ETL system deployments regardless of which tools and technologies are used. Most important — and most frequently neglected — is the practice of designing the ETL system before development begins. Too often we see systems that just evolved without any initial planning. These systems are inefficient and slow, they break down all the time, and they're unmanageable. The data warehouse team has no idea how to pinpoint the bottlenecks and problem areas of the system. A solid system design should incorporate the concepts described in detail in Kimball University: The Subsystems of ETL Revisited, by Bob Becker.
Good ETL system architects will design standard solutions to common problems such as surrogate key assignment. Excellent ETL systems will implement these standard solutions most of the time but offer enough flexibility to deviate from those standards where necessary. There are usually half a dozen ways to solve any ETL problem, and each one may be the best solution in a specific set of circumstances. Depending on your personality and fondness for solving puzzles, this can be either a blessing or a curse.
One of the rules you should try to follow is to write data as seldom as possible during the ETL process. Writing data, especially to the relational database, is one of the most expensive tasks that the ETL system performs. ETL tools contain functionality to operate on data in memory and guide the developer along a path to minimize database writes until the data is clean and ready to go into the data warehouse table. However, the relational engine is excellent at some tasks, particularly joining related data. There are times when it is more efficient to write data to a table, even index it, and let the relational engine perform a join than it is to use the ETL tool's lookup or merge operators. We usually want to use those operators, but don't overlook the powerful relational database when trying to solve a thorny performance problem.
Whether your ETL system is hand-coded or tool-based, it's your job to design the system for manageability, auditability, and restartability. Your ETL system should tag all rows in the data warehouse with some kind of batch identifier or audit key that describes exactly which process loaded the data. Your ETL system should log information about its operations, so your team can always know exactly where the process is now and how long each step is expected to take. You should build and test procedures for backing out a load, and, ideally, the system should roll back transactions in the event of a midstream failure. The best systems monitor data health during extraction and transformation, and they either improve the data or issue alerts if data quality is substandard. ETL tools can help you with the implementation of these features, but the design is up to you and your team.
Should you use an ETL tool? Yes. Do you have to use an ETL tool? No. For teams building their first or second ETL system, the main advantage of visual tools are self-documentation and a structured development path. For neophytes, these advantages are worth the cost of the tool. If you're a seasoned expert — perhaps a consultant who has built dozens of ETL systems by hand — it's tempting to stick to what has worked well in the past. With this level of expertise, you can probably build a system that performs as well, operates as smoothly, and perhaps costs less to develop than a tool-based ETL system. But many seasoned experts are consultants, so you should think objectively about how maintainable and extensible a hand-crafted ETL system might be once the consultant has moved on.
Don't expect to reap a positive return on investment in an ETL tool during the development of your first system. The advantages will come as that first phase moves into operation, as it's modified over time, and as your data warehouse grows with the addition of new business process models and associated ETL systems.
SOURCE:http://www.informationweek.com/news/software/bi/showArticle.jhtml?articleID=207002081&pgno=3&queryText=&isPrev=