You are here
A Framework for Thermal and Performance Management
Davide Basilio Bartolini, Politecnico di Milano; Filippo Sironi, Massachusetts Institute of Technology; Martina Maggio, Lund University; Riccardo Cattaneo and Donatella Sciuto, Politecnico di Milano; Marco Domenico Santambrogio, Politecnico di Milano and Massachusetts Institute of Technology
In modern computing facilities, higher and higher operating temperatures are due to the employment of power-hungry devices, hence the need for cost-effective heat dissipation solutions to guarantee proper operating temperatures. Within this context, dynamic thermal management techniques (DTM) can be highly beneficial in proactively control heat dissipation, avoiding overheating. The large-scale adoption of DTM may eventually allow the use of more cost-effective heat dissipation system, with great power consumption advantages for large datacenters.
Preventive thermal management is a technique to achieve long-term thermal control via performance degradation. However, this may result in impaired Quality of Service (QoS) and Service Level Agreements (SLAs) breaking. We address this problem by proposing a self-adaptive framework combining performance and thermal management targeting Chip Multi-Processors (CMPs). The proposed methodology harnesses control-theoretical controllers for driving idle-cycle injection and threads priority adjustment, in order to provide control over the processor temperature, while taking applications’ QoS (in terms of performance) into account. We implemented our framework in the FreeBSD operating system and evaluated it on real hardware, also comparing it with a previous framework for preventive DPTM.
Open Access Media
USENIX is committed to Open Access to the research presented at our events. Papers and proceedings are freely available to everyone once the event begins. Any video, audio, and/or slides that are posted after the event are also free and open to everyone. Support USENIX and our commitment to Open Access.