Capacity Planning with SQL Anywhere
We present an overview of issues to consider when designing a performance evaluation of SQL Anywhere. In particular, we outline potential pitfalls when designing a benchmark based on a specific database application. We address issues related both to the construction of a model database instance, and a model application workload, and outline an approach to construct a model of performance to determine the relative importance of various performance factors, such as database cache size and the server multiprogramming level. The paper contains several examples illustrating the potential of negative scalability that can occur when executing concurrent requests in an smp environment, and how these scenarios can be generated easily by artificial contention caused by poorly-defined workloads.