National Technical University of Athens (NTUA)
Μεθοδολογία παραμετροποίησης εφαρμογών βασισμένων σε ταυτόχρονες δομές δεδομένων για ενσωματωμένα συστήματα
Abstract
dc:descriptionModern embedded system architectures integrate multiple cores and they often execute applications that rely on concurrent data structures. The selection of an efficient concurrent data structure implementation is a difficult and time consuming task. Selecting a non-efficient implementation usually has negative impact on various embedded system metrics such as the performance and the energy consumption. In the present thesis, we propose a systematic methodology for the selection of effective concurrent data structures in modern embedded system applications. The methodology is semi-automatic and it is based on the exploration of the design space of concurrent data structures. It is supported by a tool flow which automates a number of steps of the methodology. The methodology allows the exploration of a number of concurrent data structures implementations by developers and the selection of the most efficient one in each context. It is applied in a number of benchmarks implemented in two modern embedded devices with different architectural specifications. The results show that by changing concurrent data structure implementations, trade-offs can be identified in many cases between metrics such as performance, energy consumption and fairness of the data structure. A similar customization methodology is proposed for the implementation of the streaming aggregation operator in embedded devices. Trade-offs between throughput, latency, required memory size and energy consumption can be identified, by selecting different customized implementations.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2016
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Papadopoulos, Lazaros
- Παπαδόπουλος, Λάζαρος
Subjects
dc:subject × 8Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/38650