In-Depth

What to look for in a BI architecture

• By Wayne W. Eckerson
• September 29, 2003

Driven to reduce the cost of maintaining a hodge-podge of business intelligence (BI) tools in multiple departments, and to increase the number of knowledge workers using BI tools, many firms are trying to standardize on an enterprise BI tools platform. When selecting an enterprise BI platform, firms need to pay attention to architectural issues. A BI toolset must have a strong architectural foundation to support enterprise processing requirements. It is pointless to deploy BI tools to thousands of staff, customers and suppliers if performance and reliability falter under peak loads.

So, what should you look for?

1. A Service-Oriented Architecture (SOA) creates shared “services” or components that encapsulate business logic inside a high-level interface. Services aren’t tied to specific apps, languages, platforms or networks. They can be invoked by any app that understands the IDL, providing a high degree of abstraction, and platform and language independence. Administrators can create multiple instances of a service on one machine, increasing availability and scalability via load balancing.

The latest industry-standard SOA frameworks are Web services (SOAP, XML and WSDL), .NET and J2EE. Most leading BI vendors have re-architected their products to support SOA frameworks and concepts. Business Objects uses CORBA. MicroStrategy and Crystal Decisions (which has been acquired by Business Objects) each adhere to Microsoft’s COM using mostly C++ components. Cognos is migrating to a new Web services architecture (SOAP and XML) that is evident in its ReportNet product.

2. Common frameworks across all products. Vendors that have fully implemented a SOA provide seamless integration among their products on many levels. Their products all share a GUI. More importantly, they run off the same object model and runtime meta data repository.

Today, only MicroStrategy offers common frameworks in these areas. Most vendors have been victimized by acquisitions. Each acquired product comes with a different interface, object model and meta data. Vendors must re-code the guts of a product while maintaining the existing code base. To minimize risks, many vendors initially hide the incompatibilities by linking them into a common portal and providing shared security until they can undertake a wholesale rewrite. Cognos, for example, has taken the latter approach.

3. Designed for server- and Web-centric processing. Most BI products have gone through the process of migrating from desktop to server-based to Web-based processing. Some haven’t completed the transition, but have made strides. For example, Business Objects, Cognos and Crystal Decisions have all released Web-based report authoring tools. MicroStrategy, which has long embraced the Web, provides nearly 100% Windows–Web functionality.

Some vendors offer plug-ins or applets to mimic desktop functionality within a Web browser. This can be a compromise between the speed and functionality of desktop processing, and the universality and ease-of-use of a Web browser. But some firms ban applets and plug-ins for security reasons.

4. Leverage multiple tiers appropriately. The best architectures perform processing where it makes the most sense, but this has been a challenge for some vendors. Many with desktop origins prefer to do as much processing on the desktop as possible. This is fine for some functions, but not for CPU-intensive functions.

MicroStrategy has touted its ROLAP architecture, which did all processing within a relational database by generating native SQL. Last year, the firm backed off a pure ROLAP approach with its “Intelligent Cube” architecture that does some processing on a mid-tier server to boost performance in large-scale environments.

5. Solid systems engineering. In delivering a scalable, high-performance BI architecture, the devil is in the systems engineering details. A good BI architecture provides sophisticated use of memory, CPU processing, storage, caching, clustering, load balancing, database pooling and session management. The architecture should also run on Windows and Unix. Some vendors are awaiting 64-bit architectures to expand their ability to cache data in memory and speed processing.

Reporting and OLAP engines require specific types of engineering to ensure good performance. Report servers need to support bursting, streaming and paging. OLAP engines must support a multi-cube architecture and fast calculations to speed load times and ensure scalability.

6. Leverage modern architecture standards. The best BI architectures leverage the latest standards and technology, but it is difficult to re-architect a product, so it’s sometimes good to be a newcomer. Informatica has built its PowerAnalyzer tool on top of Web app servers like WebSphere, which provide a robust platform for building and supporting Web apps. These servers handle core infrastructure services, like security, clustering, fault tolerance, session management, database access and portals. Most other vendors work with Web app servers, but don’t take full advantage of these infrastructure services since they were forced to build those services into their BI platforms prior to the advent of the specialized Web servers.

It’s also important for vendors to support emerging BI standards, such as MDX and ODBO for OLAP processing, and the Common Warehouse Metamodel for meta data management and interchange.

Although it’s easy to tell which platforms a product supports and whether it uses applets or DHTML to support Web-based report authoring, it is hard to evaluate a BI architecture unless you test it in a production environment. The only way to do that is to require the vendor to create a prototype app using your data and queries and then stress test the prototype using a load-testing program.

Some vendors may balk at this since it is time-consuming and expensive. But making them do it will tell you how badly they want your business and how well they are likely to support you.

Please see the following related stories:

“BI: real time or right time?” by Jack Vaughan

“Real-time light, real-time anywhere” by Alan Earls

“What’s real at Fleet?” by Jack Vaughan