Aristotle University Of Thessaloniki (AUTH)
Προσαρμοστικός έλεγχος συμφόρησης και δικαιοσύνης σε δίκτυα υπολογιστών
Abstract
dc:descriptionThe expansion and growth of traffic in packet-switching networks, have established congestion control as a highly significant unit within network design. The congestion control mechanisms in the Internet, consist of a source algorithm that dynamically adjusts source rates and a link algorithm that updates a certain congestion measure at each link and feeds it back to the sources. To solve the issue, Reno which adapts heuristically to packet loss probability, was initially proposed. Simulations studies have demonstrated that it may be ill-suited for networks with large bandwidth-delay product. Research on the theoretical understanding of TCP congestion control, lead to the development of primal - dual algorithms, analyzed via an optimization based framework, which generalize the Additive Increase - Multiplicative Decrease congestion avoidance strategy, to large-scale networks. The introduction of new types of Internet services that require strict quality criteria, silently imply the existence of control algorithms capable of regulating the round trip time, avoiding either overflow or empty queues, while guaranteing fairness among sources. Control theoretic approaches implemented as congestion controllers in the routers, may be found in where local stability around an equilibrium is addressed, since linearized versions of the fluid flow model and queue dynamics are considered. Furthermore, owing to their implementation on the links, the total number of computations performed due to congestion control, should be kept to a minimum. The practical assumption that link algorithms are unavailable to the designer, lead alternatively to the design of congestion controllers operating in the source. However, certain key questions need to be carefully addressed, prior to implementation. Specifically: a) a model for the plant has to be devised, otherwise stability analysis is impossible; b) determining a desired and feasible system performance is a non-trivial task, owing to the highly uncertain and dynamically varying nature of the Internet; c) the controlled system should not be only stable, but additionally robustly stable against modeling imperfections, exogenous disturbances, delays and errors in interpreting the network produced congestion signals; d) the output of any proposed control scheme should be saturated. Fortunately, in the source control theoretic approaches, the simplicity constraint is relaxed. The thesis proposes a Neural Network Rate Control theoretic framework, addressing all four aforementioned key questions. It was comprised: i) an on-line desired round trip estimator, whose main objective was to output reliable, piecewise constant predictions of feasible round trip times; ii) an adaptive and saturated transmission rate controller, capable of robustly tracking the desired round trip time; iii) an adaptive and saturated number of communication channels controller, capable of achieving fairness in allocating network resources among competing sources at equilibrium; iv) a future path congestion level estimator. The notion of parallel operating communication channels was introduced in order to improve the throughput of the network. It was clearly shown that regulating round trip time to follow a piecewise constant profile, is equivalent to queueing delay and queue length regulation around a piecewise constant profile. Moreover, the resulted scheme still achieves small enough queues at equilibrium, to prevent large queueing delays and congestion collapse, though not zero to avoid link starvation.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2010
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Houmkozlis, Christos
- Χούμκοζλης, Χρήστος
Subjects
dc:subject × 20- Προσαρμοστικός έλεγχος
- Δίκτυα, Νευρωνικά
- Συμφόρηση
- Δικαιοσύνη
- Δίκτυα υπολογιστών
- Πρωτόκολλο μεταφοράς
- Ευστάθεια
- Ελεγκτής ρυθμού
- Adaptive control
- Networks, Neural
- Congestion
- Fairness
- Computer networks
- Transfer protocol
- Stability
- Rate control
- 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/23929