SEER: Estimating the sampling size in performance models considering errors

Generating a performance model to optimize or to reason about variability-intensive systems is costly and the user never knows (unless she or he is a seer) when to stop sampling. The performance measurement can take from several seconds to several days, depending on the measurement and the noise, so it is very important to know if new measurements are profit. Here is where SEER can support users, by estimating the sampling size to generate performance models below a relative prediction error.

There are three folders their files that can help you to start using SEER. The folder CODES include all the source code needed to use it. SEER has been validated with ten configurable systems. Folders Ground Truth and Results include the measurements done to validate SEER.