Designing applications for heterogeneous systems, like Multiprocessor Systems on Chip with Field Programmable Gate Arrays is a complex task. In order to exploit all the capabilities of these systems, such as Partial Dynamic Reconfiguration and hardware acceleration, the designer still has to develop large parts of the system unassisted, establishing the design choices mostly on his experience. In this work we present a Mixed Integer Linear Programming formulation for mapping and scheduling of applications on heterogeneous and reconfigurable devices taking into account partial dynamic reconfiguration, module reuse and reconfiguration prefetching. Since the better these techniques and features are exploited, the better is the resulting schedule, and taking into account that the space of the solutions is considerable, it is necessary the use of automatic tools in order to help the system designer in building the best possible application design with respect to his needs, which often are not limited to improve the overall execution time (like most of the other schedulers in the literature do), but to also consider the peak power and energy consumption of the design. This means addressing a specific Resource Constrained Project Scheduling problem, which takes into consideration the reconfiguration aspects with a focus on power and energy metrics that are becoming crucial in designing applications targeting heterogeneous architectures.
Produrre applicazioni per sistemi eterogenei come i Multiprocessor System on Chip con Field Programmable Gate Array è ad oggi un compito arduo. Per sfruttare tutte le potenzialità che questi sistemi offrono, come, ad esempio, la Partial Dynamic Reconfiguration e, in generale, l’accelerazione in hardware di algoritmi, il progettista deve tuttora sviluppare considerevoli parti del sistema senza alcuna assistenza, basandosi prevalentemente sulla propria esperienza. In questa lavoro è stato sviluppato un modello di Mixed Integer Linear Programming per il mapping e lo scheduling di applicazioni su piattaforme eterogenee e riconfigurabili. Questo modello tiene in considerazione le ultime tecniche disponibili nei moderni FPGA come la partial dynamic reconfiguration, il module reuse e la reconfiguration prefetching. Per sfruttare tali tecniche al meglio, e quindi ottenere uno schedule migliore, e tenendo presente che lo spazio delle soluzioni del problema che stiamo affrontando è considerevole, si rende necessario l’utilizzo di strumenti automatici per aiutare il progettista dell’applicazione eterogenea nello sviluppo della stessa. Infatti, sempre più spesso, durante lo sviluppo di applicazioni eterogenee, non si vuole più migliorare solo il tempo di esecuzione di queste (possibilità offerta dalla maggior parte degli scheduler presenti in letteratura), ma anche la potenza a istantanea e l’energia consumata. Stiamo quindi affrontando una variante del problema di Resource Constrained Project Scheduling che deve tenere in considerazione non solo gli aspetti di riconfigurazione, ma anche metriche come potenza ed energia dello schedule.
Multiobjective reconfiguration-aware scheduling on FPGA-based heterogeneous architectures
DEIANA, ENRICO ARMENIO
2014/2015
Abstract
Designing applications for heterogeneous systems, like Multiprocessor Systems on Chip with Field Programmable Gate Arrays is a complex task. In order to exploit all the capabilities of these systems, such as Partial Dynamic Reconfiguration and hardware acceleration, the designer still has to develop large parts of the system unassisted, establishing the design choices mostly on his experience. In this work we present a Mixed Integer Linear Programming formulation for mapping and scheduling of applications on heterogeneous and reconfigurable devices taking into account partial dynamic reconfiguration, module reuse and reconfiguration prefetching. Since the better these techniques and features are exploited, the better is the resulting schedule, and taking into account that the space of the solutions is considerable, it is necessary the use of automatic tools in order to help the system designer in building the best possible application design with respect to his needs, which often are not limited to improve the overall execution time (like most of the other schedulers in the literature do), but to also consider the peak power and energy consumption of the design. This means addressing a specific Resource Constrained Project Scheduling problem, which takes into consideration the reconfiguration aspects with a focus on power and energy metrics that are becoming crucial in designing applications targeting heterogeneous architectures.File | Dimensione | Formato | |
---|---|---|---|
thesis.pdf
accessibile in internet per tutti
Descrizione: Thesis text
Dimensione
725.53 kB
Formato
Adobe PDF
|
725.53 kB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/108662