Finding optimal configurations for Stream Processing Systems (SPS) is achallenging problem due to the large number of parameters that can influencetheir performance and the lack of analytical models to anticipate the effect ofa change. To tackle this issue, we consider tuning methods where anexperimenter is given a limited budget of experiments and needs to carefullyallocate this budget to find optimal configurations. We propose in this settingBayesian Optimization for Configuration Optimization (BO4CO), an auto-tuningalgorithm that leverages Gaussian Processes (GPs) to iteratively captureposterior distributions of the configuration spaces and sequentially drive theexperimentation. Validation based on Apache Storm demonstrates that ourapproach locates optimal configurations within a limited experimental budget,with an improvement of SPS performance typically of at least an order ofmagnitude compared to existing configuration algorithms.
展开▼
