THE ROLE OF TPC-D BENCHMARK RESULTS IN SELECTING A SERVER FOR BUSINESS INTELLIGENCE: PART II
by Daniel Graham

SKILLS

Under the heading of "skills" we take a client through local vendor support. Is he proposing to install a system which local consultants can service, or will he have to fly somebody half way across the continent? How about his own staff skills and education? Bear in mind that when you talk about education it depends which server is running what database engine.

I'm with IBM, so I have a clear idea of what happens when you pair, say, an RS/6000 running Oracle. But if the customer wants to team up a more unusual combination -- perhaps because he got great performance out of his existing system -- then you owe it to him to say: wait a minute! What's your local vendor support and skill-set for this given combination?

No matter what the customer wants to combine, you are going to take a critical look at the local ISV partners in his vicinity. The customer intends to do business intelligence, including data mining and data warehousing. So you look hard at what ISVs in his area can provide in the way of general purpose tools, query and reporting, online analytical processing, extraction and cleansing, data mining, and so on.

PERFORMANCE

Next comes "performance." Here, by the way, number one on the list is TPC-D -- which brings us back to where we started. Number two is optimizer quality. Three is scalability. It is more usual for "scalability" to refer to scaling up an individual problem. Here we are talking about how many users the system can manage concurrently. That's a very different criterion from saying that the system optimizes well.

Personally, having worked through the process from concept to contract with a number of high end clients, I put skills at the top of the list. No matter what database and box the client ends up running, having access to people with the appropriate skills will help him to code around the humps; the right people will know how to compromise, make changes -- and extract the performance.

I want to move on to other aspects of selecting a server, including the merits and metrics of TPC. But first, for those customers who are still unsure of how they should rate the various factors, Dr. Thomas Saaty developed a useful piece of software. Called Analytical Hierarchical Process (AHP), it leads you through three steps: 1. it creates the hierarchy of attributes which a given customer will find most important in selecting a server (in other words, what factors do I, the customer, need to take into account?); 2. it rates the relative importance of those factors; and 3. it scores alternative offerings against each other, checking each element on the hierarchy and grading the alternatives point for point. AHP has been around for about fifteen years. At IBM we used it to assess the feature functions for the AS/400 -- for which IBM won the Malcolm Baldrige Award for excellence in manufacturing.
---
For more information, see http://www.ibm.com/bi
---
The final installment in this series will appear in the next edition of D S * .