Andrea Campanella
Automatic measurement of CPU performance under critical system conditions.
Rel. Maurizio Martina, Christian Pilato. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Elettronica (Electronic Engineering), 2021
Abstract: |
The design and implementation of a microprocessor architecture is a laborious activity, requiring the collaboration of hundreds of engineers and the use of dozens of development tools. During development, the implementation and testing phases are highly interconnected and interrelated. Each change made by the design team is followed by thorough testing of new functionality by the performance verification team. Effective communication between these two teams is crucial to meeting the tight timelines. Moreover, the more accurate the description of the bug found, the faster the identification of the cause and therefore the resolution of the problem. In this work, a solution for both of these aspects is proposed: an automatic performance capture mechanism to overcome the problem of promptness of the results and a new measurement system with more observation points in the device chain in order to obtain a more accurate picture. The results show that through a more accurate measurement and proper data presentation, debugging time can be reduced and bugs can be fixed earlier. In addition, several aspects of the performance testing phase are covered, illustrating the main benchmarking suites, the metrics used and the performance tools. The results of some significant testbench executions are then reported, with the presentation of considerations regarding the variation of program parameters and system settings. The focus is then on the implementation of the most used tool, the CHI monitor, which captures CHI protocol events on the communication bus between two devices. Through this device, it is possible to measure latency and bandwidth of the system and calculate statistics about the operations performed. The newly designed measurement system employs four CHI monitors in the system, wisely placed to extract as much useful data as possible, and replaces the old system that employed only two. Data mining from the CHI monitor log is then automated through the integration of three analysis tools into one. |
---|---|
Relators: | Maurizio Martina, Christian Pilato |
Academic year: | 2021/22 |
Publication type: | Electronic |
Number of Pages: | 93 |
Additional Information: | Tesi secretata. Fulltext non presente |
Subjects: | |
Corso di laurea: | Corso di laurea magistrale in Ingegneria Elettronica (Electronic Engineering) |
Classe di laurea: | New organization > Master science > LM-29 - ELECTRONIC ENGINEERING |
Aziende collaboratrici: | arm |
URI: | http://webthesis.biblio.polito.it/id/eprint/21183 |
Modify record (reserved for operators) |