Using event-B and Modelica to evaluate thermal management strategies in many core systems

Abstract

Dynamic thermal management is an increasingly critical and complex part of the run-time management of manycore systems. Methods of controlling temperature include thread migration, dynamic voltage and frequency scaling and power gating using various strategies and combinations of each. In the PRiME project we are developing run-time management systems to sustain the scaling of many-core systems. As part of this development we are investigating the relative benefits of different thermal management strategies by co-simulating a Modellica model of the characteristics of a many-core device with a discrete Event-B model of the run-time manager. The results enable us to efficiently design more elaborate experiments on real hardware platforms in order to validate the run time management.