It's a rare day that I concede defeat but today was one of those rare days. I threw in the towel on migrating existing Cognos Report Studio reports from a Framework Manager model that used a SQL Server relational data source to a new model based on a SQL Server Analysis Services 2005 data source.
There were two major reasons:
1. The inability to make any modifications to a Framework Manager model using objects from the SSAS cubes. This includes the creation of filters based on an SSAS cube.
2. Performance issues related to this architecture.
Let me explain why I am using a hybrid (Microsoft/Cognos) architecture . . .
A couple of years ago I took on a job to create a reporting solution for a large CRM implementation at a financial services firm. What I found was that the CRM solution that they were having customized had a plug-in module for Cognos-based "analytics". What this meant was that they had a Cognos model built on top of their OLTP system and some reports to go with it. Additionally, they provided an authentication module that allowed for users that were authenticated on the CRM system (using a combination of AD and forms based login) to have their credentials passed on to this CRM application. This was important because some of the users were on a different domain and were not able to authenticate using AD. I tried to sell them on creating a custom authentication provider for SSRS and scrap Cognos. Not because I refuse to work with non-Microsoft technology but because Cognos didn't seem like the right solution for the problem (I won't get into that here).
I couldn't sell them so I proceeded to build Cognos reports using Report Studio, against a Framework Manager model that was on top of the live OLTP database. The OLTP schema was complex, the model was buggy and full of gaps. It was not pretty. In the end we had 30+ reports. Some of which were very complex and written against a very complex and cumbersome model. The queries being produced by Cognos were unbelievably complex and at times rather difficult to tune and troubleshoot.
Fast forward a couple of years and the maker of the CRM software is now using SQL Server as their reporting and analytics platform and my client is struggling with supporting their existing Cognos implementation as they don't quite have the expertise to extend (modify existing and create new reports) or tune their solution and they would like to incorporate cubes.
Initially I thought I would build them a data mart. From that I would create a relational Cognos model as well as build out an SSAS database and cube(s). Then I thought, wouldn't it be great if I could simply build the cubes and have Framework Manager, Report Studio, and Analysis Studio use the cube(s) instead of having to maintain both the cubes and the relational model.
I got as far as building the mart, ETL, configuring an SSAS 2005 data source and creating a model on the SSAS cube. Then I began to port the existing reports. Well these reports used some filters in complex filters. I attempted to recreate these filters in the new model. No can do. It simply isn't supported. Then I started playing around with basic reports using the new model. At times the report was unresponsive. I then began looking around the message boards only to find that there were quite a number of people attempting to resolve performance issues with Cognos 8.3 and Analysis Services cubes with no positive responses. This was the point at which I threw in the towel.
So tomorrow I move on. I have some nice cubes created that will be used with Excel 2007 and hopefully a little bit with Analysis Studio for those that can't authenticate using AD. But, I'll have to define a relational model for the report conversion process.
An alternative would be to implement SSRS and recreate all the reports as well as create a custom authentication extension but that would be too costly at this time. Another alternative would be to create Cognos Transformer cubes but the client does not have a license for this product and do not want to pay for it, nor do they have the expertise to support it.
It's not an ideal solution (a relational FM model and AS cubes) and will require additional maintenance (two metadata models) but it will provide a much higher performing system with analytics capabilities and a much simpler and higher performing model from which to write reports.