Event-Based Information Fusion for the Self-Adaptive Cloud

Successful self-adaptive resource provisioning in the cloud relies on accurate tracking of workload variations and timely detection of changes in the infrastructure. The general estimation problem is very challenging due to the massive number of observable events in various subsystems, each containing some useful information. In this project, we will develop novel, event-based estimation techniques for information fusion in cloud server systems. Our starting point will be the family of Monte Carlo-based inference methods known as Particle Filters, which will be adapted to handle event-based measurements from different sources and with different time scales. The results will enable more responsive and exact decision making in the autonomous cloud.

During the second half of the project lifespan, we shifted focus towards identification of networks of microservices using queueing network models and measurements from Cloud applications. The main scientific innovation was a new mean-field fluid model for mixed queuing networks, with a smoothing parameter that can be estimated from logged timing data. The model was applied to a real microservice application and was shown to better predict queue lengths and response-time distributions than previous methods. The results were summarized in Johan Ruuskanen's doctoral thesis, which was defended on November 18.