Accurate Energy Modelling for Many-Core Static Schedules

Simon Holmbacka, Jörg Keller, Patrick Eitschberger, Johan Lilius

Tutkimustuotos: Artikkeli kirjassa/raportissa/konferenssijulkaisussaKonferenssiartikkeliTieteellinenvertaisarvioitu

5 Sitaatiot (Scopus)


Static schedules can be a preferable alternative forapplications with timing requirements and predictable behaviorsince the processing resources can be more precisely allocatedfor the given workload. Unused resources are handled by powermanagement systems to either scale down or shut off parts ofthe chip to save energy. In order to efficiently implement powermanagement, especially in many-core systems, an accurate modelis important in order to make the appropriate power managementdecisions at the right time. For making correct decisions, practicalissues such as latency for controlling the power saving techniquesshould be considered when deriving the system model, especiallyfor fine timing granularity. In this paper we present an accurateenergy model for many-core systems which includes switchinglatency of modern power saving techniques. The model is usedwhen calculating an optimal static schedule for many-core taskexecution on systems with dynamic frequency levels and sleepstate mechanisms. We create the model parameters for anembedded processor, and we validate it in practice with syntheticbenchmarks on real hardware.
AlkuperäiskieliEi tiedossa
OtsikkoParallel, Distributed and Network-Based Processing (PDP), 2015 23rd Euromicro International Conference on
Toimittajat O'Conner, Lisa
ISBN (painettu)978-1-4799-8492-3
DOI - pysyväislinkit
TilaJulkaistu - 2015
OKM-julkaisutyyppiA4 Artikkeli konferenssijulkaisuussa
TapahtumaEuromicro International Conference on Parallel, Distributed and Network-Based Processing (PDP) - 23rd Euromicro International Conference on Parallel, Distributed and Network-Based Processing (PDP 2015)
Kesto: 4 maalisk. 20156 maalisk. 2015


KonferenssiEuromicro International Conference on Parallel, Distributed and Network-Based Processing (PDP)