Athens University Economics and Business (AUEB)
Congestion control algorithms for background data transfers of minimal delay impact
Abstract
dc:descriptionThe thesis introduces congestion control algorithms for background data transfers which account for their impact on the average delay of the coexisting interactive traffic. The latter refers to flows associated with the transmission of delay-sensitive data, for which the instantaneous throughput is important. The former are indifferent to temporal throughput variations, provided some average performance guarantee. The model includes also long-lived (persistent) volume flows which (as the interactive) are out of the designer’s control. The proposed algorithms are motivated by means of two optimization problems, both capturing the pre-described time diversity. The resulting capacity allocations provide suitable incentive to background users for their adoption.The first considers long-lived background flows such as those for the Operating Systems updates. It determines their long-term throughput, while accounting the negative externalities of this magnitude on the average download times of interactive traffic. Two objective variations are formulated, which guarantee for the background users equal average throughput with the incumbent TCP, in the case of capacity scarcity or abundance respectively. Additionally, two implementation alternatives are explored for the optimal solution identification. The second considers dynamically arriving requests for background micro-files, modelling the case of cache updates. It aims to minimize the average delay of interactive traffic, while providing the desired average throughput for the background that serves the arriving demand. A set of optimal solutions is presented, under various assumptions on the level of available information and the considered network topology. The corresponding application-level congestion control algorithms implement the optimal solutions based only on end-to-end measurements and achieve significant improvement of the interactive delay. A common attribute of great importance for all algorithms, is the fact that they do not require any modifications on either the network or the transport layer of the existing internet infrastructure. For their theoretical comparative analysis, a Markovian methodology is followed. Flow-level simulated experiments provide further insights of their performance with respect to multiple parameters of the traffic mixture. Finally, packet-level experiments capture their performance in more detail, validate the theoretical results and evaluate the approaches proposed by related work.
Degree
thesis:*- Grantor dc:publisher
- Athens University Economics and Business (AUEB)
- Year dc:date
- 2016
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kanakakis, Michail
- Κανακάκης, Μιχαήλ
Subjects
dc:subject × 10- Αλγόριθμοι ελέγχου συμφόρησης
- Βέλτιστη μετάδοση δεδομένων φόντου
- Ελαχιστοποίηση της καθυστέρησης μετάδοσης δεδομένων πραγματικού χρόνου
- Congestion control algorithms
- Optimal background data transfer
- Minimization of the interactive delay
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/39950