One size fits all? - Part 2: Benchmarking results

Two years ago, some of us wrote a paper predicting the demise of "One Size Fits All (OSFA)" [Sto05a]. In that paper, we examined the stream processing and data warehouse markets and gave reasons for a substantial performance advantage to specialized architectures in both markets. Herein, we make three additional contributions. First, we present reasons why the same performance advantage is enjoyed by specialized implementations in the text processing market. Second, the major contribution of the paper is to show "apples to apples" performance numbers between commercial implementations of specialized architectures and relational DBMSs in both stream processing and data warehouses. Finally, we also show comparison numbers between an academic prototype of a specialized architecture for scientific and intelligence applications, a relational DBMS, and a widely used mathematical computation tool. In summary, there appear to be at least four markets where specialized architectures enjoy an overwhelming performance advantage.