Service-based software architectures are often modeled with queues and queuing networks. Such models are useful for performance evaluation and design. They can also assist in runtime maintenance and administration, but, in this context, it is often far more valuable to be able to forecast how QoS characteristics are likely to evolve in the near future. This is particularly important in cases where systems can be adapted to counter QoS constraint violations: in such systems, given predictions of likely future QoS characteristics, pre-emptive adaptation strategies can be implemented.
This paper outlines an approach to runtime prediction of QoS characteristics of queued systems. Predictions are computed by applying ARIMA forecasting techniques to basic properties of a queued model, and then using the model to predict complex QoS characteristics. We outline how our methods integrate into our implementation framework for monitoring and pre-emptive adaptation of web service based systems.