National Technical University of Athens (NTUA)
Παρακολούθηση και διαχείριση προσαρμογής κατανεμημένων εφαρμογών σε νεφουπολογιστικά περιβάλλοντα πολλών παρόχων
Abstract
dc:descriptionThe present work for the final evaluation of the PhD Thesis focusses on innovative research areas that have to do with distributed applications hosted in various distributed infrastructures of many cloud providers. After extensive research investigation, there are four basic research questions that finally came up as the main theme of the present PhD Thesis. Time wise, and up to intermediate evaluation of PhD thesis that took place, the first two questions that concern the dynamic monitoring of distributed applications in multi-cloud environments and the optimum reconfiguration of the infrastructure of the resources have been thoroughly investigated and analyzed. Apart from those, a third question that concerns the quite bigger accuracy of prediction in resources demand in multi-cloud environment by using federated learning techniques have been analyzed and presented during the final evaluation of PhD Thesis. Moreover, a fourth question regarding the calculation of prediction accuracy regarding the demand at the edge of the cloud with related resources restrictions is a case that has presented during the final evaluation of PhD Thesis. As per the above, in the current PhD thesis specific solutions are proposed based on the distributed and multi-layer complex event systems for effective monitoring of the infrastructure. Moreover, there is also an innovative algorithm that is proposed for optimum resources management of the multicloud environment by focusing on the optimum operation of the cloud hosted distributed applications. Finally, as part of the specific PhD thesis, two more solutions are proposed. The one concern analysis and implementation of distributed federated learning algorithms in multi cloud application environments with very good prediction demand accuracy results. Moreover, there is the case of algorithms running at the edge of the cloud referring to many providers that are examined and analyzed as well. By that way securely and without any disruption in their operation, the distributed applications are used. Our solution proposals are evaluated based on consumption of basic cloud resources such as RAM, CPU and Disk Usage and the stability and reliability of the i.e. monitoring system proposed. Regarding the decisioning model for the optimum reconfiguration of the cloud resources, measurements have taken place for the startup times of virtual machines of the public and private cloud providers. These measurements show stable performance for the public cloud providers. Moreover, results from experiments regarding the final values of the proposed algorithm for reconfiguration that takes into account startup times and application response times show the effectiveness of the new proposed method for cloud reconfiguration. What is more, distributed algorithms of federated learning for demand resources prediction show extremely good results in terms of prediction by using innovative client selection methods and technologies inspecting the adequate data volume for the training of the algorithms in multi cloud environments. Last but not least, very good results regarding the prediction accuracy of Federated Learning Architecture at the cloud edge using resources constraints and quantization methods (Tiny ML) have been successfully presented.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Stefanidis, Vasileios-Angelos
- Στεφανίδης, Βασίλειος-Άγγελος
Subjects
dc:subject × 13- Υπολογιστικό νέφος
- Επεξεργασία σύνθετων συμβάντων
- Άκρο νέφους
- Federated learning
- Cloud computing
- Complex event processing
- Cloud edge
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Υπολογιστές, Υλικό (hardware) και Αρχιτεκτονική
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Computer science, Hardware and Architecture
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/57409