Demonstration: Predicting Distributions of Service Metrics

The ability to predict conditional distributions of service metrics is key to understanding end-to-end service behavior. From conditional distributions, other metrics can be derived, such as expected values and quantiles, which are essential for assessing SLA conformance. Our demonstrator predicts conditional distributions and derived metrics estimation in realtime, using infrastructure measurements. The distributions are modeled as Gaussian mixtures whose parameters are estimated using a mixture density network. The predictions are produced for a Video-on-Demand service that runs on a testbed at KTH.

I. Background/Concepts

Understanding and predicting the performance of telecom services is difficult due to the complexity of software systems which run these services over general-purpose platforms and operating systems. Recent approaches to performance prediction are based on statistical learning, whereby models are trained with data collected from infrastructure and services. These models are then used to predict service quality, the likelihood of component failures, etc.