This month, I'll bring together the elements from the last several articles on packaged business intelligence (PBI) by highlighting an example of how one vendor is packaging business intelligence components. As discussed in my previous columns, vendors are looking for opportunities to improve their value proposition by packaging those things that can be deemed common and applicable on a broad basis. This column will highlight the offering of a prominent business intelligence player, Sun Microsystems.
Sun has long been a provider of data warehouse servers running a DBMS as well as ETL, OLAP and other business intelligence (BI) software. Although Sun servers are compatible with almost every UNIX-based BI tool, many BI environments have encountered a variety of challenges with the highly complex implementation of BI software and the estimation of the server specification, performance, reliability and interoperability of their chosen combinations prior to in-house production implementation.
Sun's recent work in developing the reference architecture program, integrating solutions for specific marketplaces, is answering the specification and integration questions and bringing empirical data into the overall implementation picture. Sun's PBI contribution is a framework for packaging the various BI layers, which they do in their three major Sun iForce Centers.
Box specification is accomplished by comparison to successful BI environments and Sun iForce Center implementations, looking at factors of data size, number of users, complexity of queries and batch availabilities to determine a shop's uniqueness from the reference accounts and determine necessary customizations.
Current reference architectures include integration of components from the DBMS, ETL and business performance management spaces including products from Oracle, Sybase, Informatica and Hyperion. Based on reference points from a variety of implementations and tests in the centers, Sun has compiled performance numbers from a variety of common data warehouse queries, data warehouse loads, multidimensional cube loads and multidimensional cube queries.
Production implementation considerations such as concurrency affect performance results; and Sun, through their technical documentation, helps predict the degree of the impact. Each reference architecture's performance across the range of data warehouse and cube queries and load/build times can be accurately predicted up to a high number of concurrent users. Keep in mind that concurrent users in a lab are truly concurrent (i.e., not accounting for think time) and probably account for a tenfold number of true, production concurrent users.
Example queries used are representative of BI environments and include analysis of subjects such as distribution channels, products, promotions, purchases and payments to analyze business metrics such as customer profitability, channel profitability, product profitability and promotion effectiveness.
Performance is also obviously affected by a wide range of architecture issues. Sun has recommended architecture customizations such as partitioning of the database, the automatic reallocation of the data warehouse server's CPUs based on daily usage patterns ("domaining"), cube segmentation and dimensional modeling. These are, however, valid performance tuning techniques that merit your consideration.
These architecture customizations could easily fill a few articles each to fully explore, but consider briefly cube segmentation. Cube loading times and space requirements have been issues ever since there were cubes. The performance advantage of cubes can quickly be removed by too much of a good thing adding more and more dimensions.
Hyperion Essbase part of the reference architectures has the feature of database partitioning which allows physical partitioning of the cube (e.g., by month) tied together by a virtual cube layer which keeps the access layer intact. As in a partitioned relational database, the cube partition becomes the level that the load impacts.
Other performance tuning measures related to cube access include the number of dimensions, the number of hierarchies in a dimension, the number of levels in the dimensions and the number of members in a dimension. Variations of the impact on performance for these custom measures as well as machine specifications such as CPUs are also part of the reference architectures and are found in the implementation and sizing guides.
No two environments are alike. Therefore, Sun also has a framework for using the performance numbers as guidelines for any environment based on the customization of a variety of factors including data size, load volumes and machine selection. Performance is very predictable by choosing the exact components in the prebuilt reference architecture software set as well as choosing from the recommended architecture constructs. However, the Sun iForce centers can and should be used if your software selections differ from a reference architecture and you still need performance estimations.
An investment in an offering of this nature is worthwhile for new or expanding midsize and large business intelligence environments. Reference architectures take the software compatibility questions off the table and allow you to go in "heads up" on the capabilities of your system and with solid architecture guidance. Engaging the reference architectures program can also provide interesting data points for struggling or emerging BI programs that have the opportunity to swap components.
Sun's package is different from the consulting/software PBI variety in that the "programming" aspects of BI such as the data model and the ETL and OLAP development are developed by a consultancy or in house. Alternatively, the Sun reference architecture can be the foundation of other packages that include prebuilt data models, ETL programs and analytic applications.
These PBI combination possibilities highlight the new, packaged BI world with multiple buy components and custom extensions to achieve BI. The possibilities are limited only by the imagination.
Register or login for access to this item and much more
All Information Management content is archived after seven days.
Community members receive:
- All recent and archived articles
- Conference offers and updates
- A full menu of enewsletter options
- Web seminars, white papers, ebooks
Already have an account? Log In
Don't have an account? Register for Free Unlimited Access