National Technical University of Athens (NTUA)
Ποιότητα υπηρεσίας στο υπολογιστικό πλέγμα
Abstract
dc:descriptionA grid is a wide-area distributed system spanning a large number of heterogeneous resources, which are bridged by middleware. Those characteristics, combined with the generally unpredictable node availability on a loosely coupled system, render QoS assurance for those systems a highly challenging task. The problem can be logically split into three subproblems; The first subproblem is obtaining accurate metrics for supporting QoS decisions, as well as disseminating those metrics throughout the grid on time. For instance, it is desirable for a grid scheduler to know the (static and real-time) performance characteristics of the worker nodes it is managing, so as to forward high-priority jobs to the fastest nodes. The second subproblem concerns 1-1 mapping of low-level system performance metrics and qualitative characteristics suitable for dealing with resources when performing a task to/from high-level ontologies which are more suitable for signing and enforcing Service Level Agreements. The third subproblem deals with designing the middleware itself so that it supports notions inherent in QoS-aware systems, such as optimal resource use, scalability and execution migration. This thesis deals with the first and third research directions by implementing decentralized (choreography-based) solutions to concrete problems. Regarding the first research direction, the thesis focuses on the problem of distance measurement in volunteer computing systems (an important class of grids). A completely decentralized algorithm with zero network overhead is proposed and tested through its implementation. Additionally, we present a generic framework for implementing simple peer-to-peer and grid systems. The framework (sp2p) is specifically tuned to research purposes, featuring a very steep learning curve and allowing rapid implementation and easy testing of experimental architectures and protocols. Regarding the third research direction, the thesis focuses on the problem of fine-grained scheduling on clusters (clusters being by far the most important class of resources operating within a grid). We implement and present a low-level middleware (Cronus) which performs fine-grained parallel execution of uniform nested loops on clusters, employing a dynamic and decentralized scheduling algorithm. Finally, we introduce Cronus/G, a grid service that “wraps” Cronus and functions as a mediation core for submitting uniform nested loops on a grid with QoS assurance.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2009
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kamenopoulos, Dimitrios
- Καμενόπουλος, Δημήτριος
Subjects
dc:subject × 14- Υπολογιστικό πλέγμα
- Ποιότητα υπηρεσιών
- Ομότιμα δίκτυα
- Ομοιόμορφοι φωλιασμένοι βρόχοι
- Αποκεντρωμένες αρχιτεκτονικές
- Computer grid
- Quality of services (QoS)
- Peer-to-peer networks
- Uniform nested loops
- Decentralized architectures
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/18292