This chapter presents the proposed event-based thermal control solution at the HARPA-RT level, thus complementing the previous chapter which dealt with the same topic at the HARPA-OS level. A brief analysis of the thermal control problem is presented, evidencing as the main requirement the need for thermal control at the millisecond timescale, caused by software variability in the use of CPU functional resources and fast thermal dynamics inside the silicon die. To meet this requirement while keeping a low overhead, the proposed solution consists of a hardware state machine and datapath that monitors temperature and generates interrupts, and a software policy built using event-based control theory. This partition provides both a fast response to critical and unpredictable temperature increases, a very low overhead when temperature is low or almost constant, and the flexibility of a software implementation of the control policy. The proposed solution is evaluated both in simulation using the Modelica modelling language, and on a commercial Intel CPU.

Event-Based Thermal Control for High Power Density Microprocessors

Federico Terraneo;Alberto Leva;William Fornaciari
2019-01-01

Abstract

This chapter presents the proposed event-based thermal control solution at the HARPA-RT level, thus complementing the previous chapter which dealt with the same topic at the HARPA-OS level. A brief analysis of the thermal control problem is presented, evidencing as the main requirement the need for thermal control at the millisecond timescale, caused by software variability in the use of CPU functional resources and fast thermal dynamics inside the silicon die. To meet this requirement while keeping a low overhead, the proposed solution consists of a hardware state machine and datapath that monitors temperature and generates interrupts, and a software policy built using event-based control theory. This partition provides both a fast response to critical and unpredictable temperature increases, a very low overhead when temperature is low or almost constant, and the flexibility of a software implementation of the control policy. The proposed solution is evaluated both in simulation using the Modelica modelling language, and on a commercial Intel CPU.
2019
Harnessing Performance Variability in Embedded and High-performance Many/Multi-core Platforms
978-3-319-91961-4
978-3-319-91962-1
Thermal management
File in questo prodotto:
File Dimensione Formato  
CH5 Event Controller.pdf

Accesso riservato

Descrizione: camera ready
: Post-Print (DRAFT o Author’s Accepted Manuscript-AAM)
Dimensione 1.37 MB
Formato Adobe PDF
1.37 MB Adobe PDF   Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11311/1066940
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 5
  • ???jsp.display-item.citation.isi??? ND
social impact