Defesa de Mestrado – Iago de Oliveira Silvestre – 29/3/2022
22/03/2022 14:00
| Defesa de Dissertação de Mestrado | |
| Aluno | Iago de Oliveira Silvestre |
| Orientador | Prof. Leandro Buss Becker, Dr. – DAS/UFSC |
| Data | 29/3/2022 (quinta-feira) – 8h30
Videoconferência (https://conferenciaweb.rnp.br/spaces/defesa-iago-silvestre) |
| Banca | Prof. Leandro Buss Becker, Dr. – DAS/UFSC (presidente);
Prof. Antonio Carlos Beck Filho, Dr. – INF/UFRGS; Prof. Giovani Gracioli, Dr. – PPGCC/UFSC; Prof. Carlos Barros Montez, Dr. – DAS/UFSC. |
| Título | Using Gem5 Simulator to Support Design Space Exploration Targeting Arm Architecture |
| Abstract: Making the performance analysis of embedded systems is very important when dealing with Cyber-Physical Systems that require stability guarantees. They typically operate having to respect temporal constraints imposed during the design of the related control system. Until recently, performance analysis was done exclusively by executing the code on the target platform and making measures. Usually code execution/measuring can also be done on simulation software, which offers greater degree of flexibility for designers to configure the platform accordingly for the desired tests. However, the usually of such simulators is a common concern. Besides, there are doubts whether the results from such simulation software are reliable enough to guide Design alternatives during a project development. The main focus of this Master Thesis is to analyse the precision of the gem5 simulation tool for simulating control algorithms. It is envisioned that obtained simulation results can help designers selecting the embedded computing architecture. It is used as case study the control algorithms developed within the ProVANT project, which serve for controlling an Unmanned Aerial Vehicle (UAV). Such analysis is important given the following reasons: (i) allows to better understand the timing behavior of the algorithms; (ii) allows evaluating architectural issues related with the embedded platform, and (iii) allows determining how accurate is the simulation software. The development board Raspberry Pi 3 Model B+ (with Cortex-A53 processor) is used as reference platform and serves as basis for creating different simulated versions for the analysis. Obtained results showed that gem5 has enough timing precision when compared with results from the real hardware, at least with respect to execution time calculation, proving to be a useful tool to explore potential hardware and software alternatives when dealing with a project using the ARM architecture. Tests comparing real hardware with the gem5 simulation did show however some mismatches on data related to lowest-level cache accesses, which in the Raspberry Pi 3 B+ is level 2. An additional contribution of this work is the creation of a cloud computing environment at Google Colab, making significantly easier the use of gem5 simulator, even allowing its usage by non-experts. | |
