|Tipo di tesi||Tesi di laurea magistrale|
|Titolo||Un approccio a livello scheduler per il controllo fine della banda di accesso in memoria in sistemi eterogenei basati su FPGA.|
|Titolo in inglese||A Scheduler-level approach for fine-grained memory bandwidth regulation in FPGA-based heterogeneous platforms.|
|Struttura||Dipartimento di Scienze Fisiche, Informatiche e Matematiche|
|Corso di studi||INFORMATICA (D.M. 270/04)|
|Data inizio appello||2022-04-13|
|Disponibilità||Accessibile via web (tutti i file della tesi sono accessibili)|
I sistemi eterogenei su singolo chip (HeSoCs) sono ampiamente sfruttati in molti domini applicativi, ed includono processori per scopi specifici accoppiati ad acceleratori come Graphics Processing Units (GPU) o Field Programmable Gate Arrays (FPGA).
Heterogeneous System-On-Chips (HeSoCs) are widely exploited in many application domains and include general-purpose CPUs coupled to accelerators like Graphics Processing Units (GPU) or Field Programmable Gate Arrays (FPGA). These systems rely on shared memory communication, and concurrent accesses from many computing engines like CPU or FPGA may hamper task execution latency and bandwidth. Solutions for Runtime Bandwidth Regulation (RDB) have been recently proposed in FPGA-based HeSoCs, that attempt to mitigate the problem of CPU/FPGA interference in main memory. RBR has been implemented and tested on a Xilinx Zynq UltraScale+ platform for bare-metal execution. This limits the applicability of the approach and calls for integrated solutions with mainstream operating systems. This thesis proposed a framework for the integration of RBR into the Linux kernel. A simple interference aware controller is proposed that hooks the linux scheduler to monitor CPU task execution and throttle FPGA tasks accordingly, with the final goal to deliver the requested Quality-of-Service (QoS). The proposed solution simplifies the adoption of RBR in FPGA-based HeSoCs. The experimental evaluation assesses the cost of the controller and its impact on the execution time of the tasks and on the mitigation of the effects of interference.