National Technical University of Athens (NTUA)
Μεθοδολογίες και εργαλεία διαχείρισης πόρων και παραμετροποίησης εφαρμογών κατά το χρόνο εκτέλεσης σε πολυπύρηνες ενσωματωμένες πλατφόρμες
Abstract
dc:descriptionIn this Ph.D. Thesis, we present (i) memory management middleware acceleration and customization______________________________________________________________________Εθνικό Κέντρο Τεκμηρίωσης | ΕΑΔΔ | 4methodologies for applying customized dynamic memory managers (allocators) and (ii) frameworks fordistributed run-time resource management on many-core platforms. Firtsly, the customization isachieved by applying, on the middleware level, custom microcoded memory allocators. Secondly, therun-time resource management on the platform is achieved by using cores in different roles and byapplying a distributed on-chip communication scheme. The proposed methodologies showed that themicrocode approach is a good alternative to overcome the performance flexibility dilemma, offering aprogrammable and flexible solution for accelerating a wide range of applications. Thus, we adopt themicrocoded approach to address memory management issues on Distributed Shared Memory (DSM)many-core embedded platforms, aiming for hardware performance but maintaining the flexibility ofprograms. Also, the developed framework provides a flexible solution in the run-time mapping problemoffering different levels of platform utilization according to application’s needs and without a centralpoint of failure. Concerning microcoded memory management services, experimental results show thatthe gain, of the proposed approaches for designing customized microcoded memory managers, wasapproximately 7x for served allocation requests with a small increase of approximately 14% to averageenergy consumption per allocation. The run-time resource management framework adapts toapplication’s needs and application’s execution restrictions by using the matching factor parameter andproduces on average 21% and 10% better on-chip communication cost for homogeneous andheterogeneous platforms respectively. Last, concerning the malleable parallel applications, thedeveloped framework has on average 70% less messages, 64% smaller message size and 20%application speed-up gain.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2014
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Anagnostopoulos, Iraklis
- Αναγνωστόπουλος, Ηρακλής
Subjects
dc:subject × 20- Δυναμική διαχείριση μνήμης
- Πολυπύρηνα ενσωματωμένα συστήματα
- Δυναμική διαχείριση τάσης και συχνότητας
- Απεικόνιση στο χρόνο εκτέλεσης
- Δίκτυο σε ψηφίδα
- Διαχείριση πόρων
- Dynamic memory management
- Many-core embedded systems
- Dynamic voltage and frequency scaling
- Run time mapping
- Network on chip
- Resource management
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Natural Sciences
- Computer and Information Sciences
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/35468